forum_id
stringlengths 8
20
| forum_title
stringlengths 4
171
| forum_authors
sequencelengths 0
25
| forum_abstract
stringlengths 4
4.27k
| forum_keywords
sequencelengths 1
10
| forum_pdf_url
stringlengths 38
50
| note_id
stringlengths 8
13
| note_type
stringclasses 6
values | note_created
int64 1,360B
1,736B
| note_replyto
stringlengths 8
20
| note_readers
sequencelengths 1
5
| note_signatures
sequencelengths 1
1
| note_text
stringlengths 10
16.6k
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|
D9ggc3l0wi | Leveraging Sparse Input and Sparse Models: Efficient Distributed Learning in Resource-Constrained Environments | [
"Emmanouil Kariotakis",
"Grigorios Tsagkatakis",
"Panagiotis Tsakalides",
"Anastasios Kyrillidis"
] | Optimizing for reduced computational and bandwidth resources enables model training in less-than-ideal environments and paves the way for practical and accessible AI solutions. This work is about the study and design of a system that exploits sparsity in the input layer and intermediate layers of a neural network. Further, the system gets trained and operates in a distributed manner. Focusing on image classification tasks, our system efficiently utilizes reduced portions of the input image data. By exploiting transfer learning techniques, it employs a pre-trained feature extractor, with the encoded representations being subsequently introduced into selected subnets of the system's final classification module, adopting the Independent Subnetwork Training (IST) algorithm. This way, the input and subsequent feedforward layers are trained via sparse ``actions'', where input and intermediate features are subsampled and propagated in the forward layers.
We conduct experiments on several benchmark datasets, including CIFAR-$10$, NWPU-RESISC$45$, and the Aerial Image dataset. The results consistently showcase appealing accuracy despite sparsity: it is surprising that, empirically, there are cases where fixed masks could potentially outperform random masks and that the model achieves comparable or even superior accuracy with only a fraction ($50\%$ or less) of the original image, making it particularly relevant in bandwidth-constrained scenarios. This further highlights the robustness of learned features extracted by ViT, offering the potential for parsimonious image data representation with sparse models in distributed learning. | [
"sparse neural network training",
"efficient training"
] | https://openreview.net/pdf?id=D9ggc3l0wi | teC36U73ZR | official_review | 1,696,620,105,254 | D9ggc3l0wi | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission46/Reviewer_WvbE"
] | title: Official Review
review: **Summary** This work studies and designs a system that exploits sparsity in the input layer as well as the intermediate layers in a resource-constrained scenario. The system uses a single masked representation of each image during training and then employs independent subnetwork training algorithm. The authors show that a single masked representation of each image can match the performance of randomly drawing masks at each training iteration.
**Pros** I must first make a disclaimer that I don't have a background in designing efficient distributed learning system, therefore, I am unable to evaluate the authors' claim on efficiency, etc. I personally find it surprising that a single masked representation of each image can work almost as well as drawing random masks at each training iteration since this is basically like making the model only see part of the image during training. And this is the key the authors use to design the system since this sparse representation can bring down the computational cost.
**Cons** With that being said, I do feel the experiment section is lacking for this work, even from a layman's perspective. Based on what I see, most of the experiments are about accuracy. If the authors are trying to claim their system is efficient, I believe experiment results on, say, training time, memory, etc should be shown. Also, it is not clear how this work is compared to previous work on efficient learning, as the authors didn't benchmark their approach to other existing methods.
rating: 5: Marginally below acceptance threshold
confidence: 3: The reviewer is fairly confident that the evaluation is correct |
D9ggc3l0wi | Leveraging Sparse Input and Sparse Models: Efficient Distributed Learning in Resource-Constrained Environments | [
"Emmanouil Kariotakis",
"Grigorios Tsagkatakis",
"Panagiotis Tsakalides",
"Anastasios Kyrillidis"
] | Optimizing for reduced computational and bandwidth resources enables model training in less-than-ideal environments and paves the way for practical and accessible AI solutions. This work is about the study and design of a system that exploits sparsity in the input layer and intermediate layers of a neural network. Further, the system gets trained and operates in a distributed manner. Focusing on image classification tasks, our system efficiently utilizes reduced portions of the input image data. By exploiting transfer learning techniques, it employs a pre-trained feature extractor, with the encoded representations being subsequently introduced into selected subnets of the system's final classification module, adopting the Independent Subnetwork Training (IST) algorithm. This way, the input and subsequent feedforward layers are trained via sparse ``actions'', where input and intermediate features are subsampled and propagated in the forward layers.
We conduct experiments on several benchmark datasets, including CIFAR-$10$, NWPU-RESISC$45$, and the Aerial Image dataset. The results consistently showcase appealing accuracy despite sparsity: it is surprising that, empirically, there are cases where fixed masks could potentially outperform random masks and that the model achieves comparable or even superior accuracy with only a fraction ($50\%$ or less) of the original image, making it particularly relevant in bandwidth-constrained scenarios. This further highlights the robustness of learned features extracted by ViT, offering the potential for parsimonious image data representation with sparse models in distributed learning. | [
"sparse neural network training",
"efficient training"
] | https://openreview.net/pdf?id=D9ggc3l0wi | ei2QDp2zjD | decision | 1,700,363,939,249 | D9ggc3l0wi | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Program_Chairs"
] | decision: Accept (Oral)
comment: The paper presents a novel approach to exploit sparsity in both the input and intermediate layers of a neural network in a resource-constrained distributed learning scenario. Reviewers find the idea innovative and acknowledge the potential impact. They appreciate the clarity of writing and the experiments conducted but raise concerns about the lack of comparisons with other efficient learning methods, the performance drop with reduced parameters, the choice of fixed masks, and the need for actual training time measurements in a distributed setting. Reviewers also suggest that the paper's tone and contributions be clarified and that experiments on training from scratch and more efficient models be considered. Despite concerns, AC and reviewers still agree that the pros outweigh the cons, and recommend acceptance.
The action PC chair for this paper is Atlas Wang, who made the decision after carefully reading the paper as well as the comments by all reviewers and AC. The decision is agreed by all PC chairs.
title: Paper Decision |
D9ggc3l0wi | Leveraging Sparse Input and Sparse Models: Efficient Distributed Learning in Resource-Constrained Environments | [
"Emmanouil Kariotakis",
"Grigorios Tsagkatakis",
"Panagiotis Tsakalides",
"Anastasios Kyrillidis"
] | Optimizing for reduced computational and bandwidth resources enables model training in less-than-ideal environments and paves the way for practical and accessible AI solutions. This work is about the study and design of a system that exploits sparsity in the input layer and intermediate layers of a neural network. Further, the system gets trained and operates in a distributed manner. Focusing on image classification tasks, our system efficiently utilizes reduced portions of the input image data. By exploiting transfer learning techniques, it employs a pre-trained feature extractor, with the encoded representations being subsequently introduced into selected subnets of the system's final classification module, adopting the Independent Subnetwork Training (IST) algorithm. This way, the input and subsequent feedforward layers are trained via sparse ``actions'', where input and intermediate features are subsampled and propagated in the forward layers.
We conduct experiments on several benchmark datasets, including CIFAR-$10$, NWPU-RESISC$45$, and the Aerial Image dataset. The results consistently showcase appealing accuracy despite sparsity: it is surprising that, empirically, there are cases where fixed masks could potentially outperform random masks and that the model achieves comparable or even superior accuracy with only a fraction ($50\%$ or less) of the original image, making it particularly relevant in bandwidth-constrained scenarios. This further highlights the robustness of learned features extracted by ViT, offering the potential for parsimonious image data representation with sparse models in distributed learning. | [
"sparse neural network training",
"efficient training"
] | https://openreview.net/pdf?id=D9ggc3l0wi | eE5PP1IGjM | official_review | 1,696,549,629,881 | D9ggc3l0wi | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission46/Reviewer_jfHR"
] | title: The findings of fixed mask pattern and training with masked image are interesting
review: **Summary**
This work studies the sparsity in the input image as well as the model parameters in vision transformers. Specifically, this paper focuses on the transfer learning from a pre-trained Masked Autoencoder (MAE) and ViT, where the backbone ViT encoder is fixed and only a MLP is trained on image classification tasks. With empirical study, it firstly finds that applying a fixed mask pattern on patches of all examples can achieve comparable or even better performance of applying random mask to different examples at different iterations. This finding suggests that dataset can be stored in reduced size by removing some patches, i.e., masked images. Secondly, this paper finds that masked images can even enhance the performance from original images. Lastly, this paper proposes to train the model in a distributed setting via Independent Subnetwork Training (IST), where a subnet of the dense net is deployed and trained on different nodes/sites. This distributed setting can further reduce the workload per worker.
**Strengths**
The findings of utilizing fixed mask pattern or random mask pattern in transfer learning is interesting, and the proposed masked image storage is also helpful in practice.
**Weaknesses**
1. Most conclusions of this paper are based on transfer learning, where the ViT backbone is fixed and only a shallow MLP is trained. Thus, whether these conclusions can generalize to training from scratch. For example, when training MAE with a fixed mask pattern, can we get a similar conclusion? Meanwhile, this paper does not mention this transfer learning setting in the abstract and summaries, but try to claim these findings are generalized conclusions. To this end, it would be better if the authors can fix the overall tone of this work.
2. The need for distributed training of the MLP layer in transfer learning is in doubt. This paper only evaluates a two-layer MLP in transfer learning, which can easily fit into any modern GPUs or even edge computing. Thus, the usage of distributed training, which splits this lightweight MLP into several subnets, may not be necessary. Thus, the contribution of this part can be challenged. It would be more meaningful if this paper shows that the proposed distributed training pipeline can be generalized to some heavy models, such as LLM and Stable Diffusions, where we even can not load the entire model into a single GPU.
3. Some contents in this paper are kind of redundant. For example, Figure 5 and Figure 6 is just the visualization of all data in Table 4. Since this paper only aims to show the linear relationship between the mask ratio and the dataset size, which can be easily concluded from the equation in Line 261. Besides, Figure 6 aims to help claim the relationship between mask ratio and performance (Line 265), but shows it with dataset storages and performance, which is kind of misleading.
rating: 4: Ok but not good enough - rejection
confidence: 4: The reviewer is confident but not absolutely certain that the evaluation is correct |
D9ggc3l0wi | Leveraging Sparse Input and Sparse Models: Efficient Distributed Learning in Resource-Constrained Environments | [
"Emmanouil Kariotakis",
"Grigorios Tsagkatakis",
"Panagiotis Tsakalides",
"Anastasios Kyrillidis"
] | Optimizing for reduced computational and bandwidth resources enables model training in less-than-ideal environments and paves the way for practical and accessible AI solutions. This work is about the study and design of a system that exploits sparsity in the input layer and intermediate layers of a neural network. Further, the system gets trained and operates in a distributed manner. Focusing on image classification tasks, our system efficiently utilizes reduced portions of the input image data. By exploiting transfer learning techniques, it employs a pre-trained feature extractor, with the encoded representations being subsequently introduced into selected subnets of the system's final classification module, adopting the Independent Subnetwork Training (IST) algorithm. This way, the input and subsequent feedforward layers are trained via sparse ``actions'', where input and intermediate features are subsampled and propagated in the forward layers.
We conduct experiments on several benchmark datasets, including CIFAR-$10$, NWPU-RESISC$45$, and the Aerial Image dataset. The results consistently showcase appealing accuracy despite sparsity: it is surprising that, empirically, there are cases where fixed masks could potentially outperform random masks and that the model achieves comparable or even superior accuracy with only a fraction ($50\%$ or less) of the original image, making it particularly relevant in bandwidth-constrained scenarios. This further highlights the robustness of learned features extracted by ViT, offering the potential for parsimonious image data representation with sparse models in distributed learning. | [
"sparse neural network training",
"efficient training"
] | https://openreview.net/pdf?id=D9ggc3l0wi | WRJk0wG3L1 | meta_review | 1,699,761,051,087 | D9ggc3l0wi | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission46/Area_Chair_b5js"
] | metareview: This manuscript presents an innovative concept of utilizing a masked input image for resource-constrained workers in a distributed learning setting. The three reviewers initially raised concerns about the lack of compelling empirical evidence to substantiate the practical benefits of implementing such sparsity. Furthermore, the training seemed to only fine-tune a two-layer MLP component for moderate models, a task manageable by any modern GPU.
However, upon discussion within our review panel, the novel contributions and the potential impact of the work become evident. The paper is centered on the optimization of computational resources and the reduction of bandwidth requirements, enabling model training even in less-than-ideal environments. The novel approach of exploiting sparsity in both the input and intermediate layers of a neural network is a valuable contribution to the field. The experimental results, as conducted on several benchmark datasets, demonstrate remarkable accuracy despite imposing sparsity. The finding that fixed masks can outperform random ones, and that the model achieves comparable or even superior accuracy with only a fraction of the original image, is particularly interesting in bandwidth-constrained scenarios.
In light of these considerations, I recommend acceptance of the paper. I strongly recommend the authors taking the reviewers' concerns into account, when preparing the camera-ready submission.
recommendation: Accept (Poster)
confidence: 4: The area chair is confident but not absolutely certain |
1AEJSYO6GX | Closed-Loop Transcription via Convolutional Sparse Coding | [
"Xili Dai",
"Ke Chen",
"Shengbang Tong",
"Jingyuan Zhang",
"Xingjian Gao",
"Mingyang Li",
"Druv Pai",
"Yuexiang Zhai",
"Xiaojun Yuan",
"Heung-Yeung Shum",
"Lionel Ni",
"Yi Ma"
] | Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets. | [
"Convolutional Sparse Coding",
"Inverse Problem",
"Closed-Loop Transcription"
] | https://openreview.net/pdf?id=1AEJSYO6GX | yhzuMV3YI9 | official_review | 1,696,699,608,129 | 1AEJSYO6GX | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission40/Reviewer_iEnQ"
] | title: Important direction and a well-written paper. The method is principled but may lack practicality.
review: Summary:
This paper proposes an auto-encoding architecture that implements convolutional sparse coding layers with weighting sharing between the encoder and the decoder. Trained with CTRL, the model exhibits decent scalability, structured representation, and superior sample-wise alignment.
Merits:
1. The paper is well written, carefully introducing and discussing technical details.
2. The direction toward more powerful white-box models is relevant and essential in the era of large black-box models.
3. The proposed approach is well-grounded with principles. Each component has a clear interpretation, including the convolutional kernels shared by the encoder, the decoder, and the CTRL objective.
4. The method exhibits stronger scalability and reconstruction performance than its precursor CTRL.
Disadvantages:
1. I am unsure about the experimental comparison with popular deep-learning generative models like VAEs and GANs. If I understand it correctly, the proposed approach can also perform reconstruction, whereas VAEs and GANs are designed to sample novel samples from the training distribution. Therefore, the comparison in Table 1 seems unfair (or not comparable) to me. Plus, the lack of the capability of sampling novel samples is undesirable in comparison.
2. The encoding process involves iterative algorithms, which may significantly slow the training and induce numerical instability. On this aspect, it would be appreciated if the authors could offer training time reports contrasting the evaluated approaches.
rating: 6: Marginally above acceptance threshold
confidence: 2: The reviewer is willing to defend the evaluation, but it is quite likely that the reviewer did not understand central parts of the paper |
1AEJSYO6GX | Closed-Loop Transcription via Convolutional Sparse Coding | [
"Xili Dai",
"Ke Chen",
"Shengbang Tong",
"Jingyuan Zhang",
"Xingjian Gao",
"Mingyang Li",
"Druv Pai",
"Yuexiang Zhai",
"Xiaojun Yuan",
"Heung-Yeung Shum",
"Lionel Ni",
"Yi Ma"
] | Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets. | [
"Convolutional Sparse Coding",
"Inverse Problem",
"Closed-Loop Transcription"
] | https://openreview.net/pdf?id=1AEJSYO6GX | Xk21TgTAlM | official_review | 1,696,653,526,964 | 1AEJSYO6GX | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission40/Reviewer_jNki"
] | title: Well-written paper with convincing experiments
review: This paper introduces a new autoencoder design that uses convolutional sparse coding (CSC) and trains it using the closed-loop transcription (CTRL) method. The resulting generative model performs well in both image reconstruction and generation.
Pros:
1. Clear writing: The paper is easy to understand, and the experiments are well explained and supported.
2. Good literature review: The related works section covers the most relevant research in the field.
3. Strong experiments: The experiments are convincing, with comparisons to other models that make the proposed approach look solid.
Cons:
1. Discuss weaknesses: The paper should discuss its limitations more clearly.
2. Higher resolution experiments: Please test the model on higher-resolution datasets like 128x128 if possible, or explain why not and how it compares to the baselines in this regard.
3. Include timing information: Please clarify how long the training and inference take compared to the baselines.
4. Reconstruction quality metrics: Please use metrics like PSNR or SSIM to measure image quality in Figures 2, 3, and 5.
5. Please show the number of trainable parameters for different models in Table 1. You can also add training and inference times to this table.
6. Please Include the original images in Figure 5 so readers can see how well the model recreates them.
7. Please make Figure 4 clearer by adding more space between blocks and brief captions for each block.
I believe this is overall a decent paper. The proposed framework is promising but should address these concerns to improve its overall quality and clarity. These improvements could lead to a higher evaluation score.
rating: 6: Marginally above acceptance threshold
confidence: 4: The reviewer is confident but not absolutely certain that the evaluation is correct |
1AEJSYO6GX | Closed-Loop Transcription via Convolutional Sparse Coding | [
"Xili Dai",
"Ke Chen",
"Shengbang Tong",
"Jingyuan Zhang",
"Xingjian Gao",
"Mingyang Li",
"Druv Pai",
"Yuexiang Zhai",
"Xiaojun Yuan",
"Heung-Yeung Shum",
"Lionel Ni",
"Yi Ma"
] | Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets. | [
"Convolutional Sparse Coding",
"Inverse Problem",
"Closed-Loop Transcription"
] | https://openreview.net/pdf?id=1AEJSYO6GX | Vqg0ejjNyz | official_review | 1,696,805,557,910 | 1AEJSYO6GX | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission40/Reviewer_BBBs"
] | title: Incorporating convolutional sparse coding in CTRL generative framework
review: **Summary of the proposed work:**
Conventional autoencoders typically use generic network architectures and lack structure or interpretability in the latent code representations. A recently proposed CTRL framework uses rate reduction to build generative models while imposing a Gaussian mixture model like structure in the latent codes.
The submitted work proposes CSC-CTRL in which the generation of natural images is modeled using a stacked concatenation of multiple convolutional sparse coding (CSC) layers and incorporated within the overall CTRL framework. The atoms in the convolutional dictionaries of the encoder and decoder are kept the same which allows for sample-wise alignment in the generated samples—an improvement over existing CTRL approach.
The paper shows good image generation quality with interpretability in latent codes with respect to image classes in the dataset. There is additionally a higher stability to noise than vanilla CTRL owing to the incorporation of sparsity.
**Pros:**
- The results are nice and convincing with respect to the main ideas in the paper’s body.
- Barring a few minor typos, the paper is generally clearly written.
**Cons:**
I have the following questions/concerns that I would like the authors to clarify.
- The authors emphasize that the key goal is to show the feasibility of high quality image generation using CSC layers. While their results are interesting, there exist a few recent works which also use multi-scale CSC to generate high quality images, for example [1] and [2]. These focus on imaging inverse problems rather than autoencoding, though one could extend them to autoencoding by say tying the weights of their encoders and decoders.
There are indeed some differences between the proposed work and the 2 references, for example, [1] performs direct pixel level MSE loss minimization instead of distribution matching or CTRL. However, given similarities like CSC based high quality generation and interpretability in obtained dictionaries, a comparison (conceptual or experimental) would be helpful to better place the contributions of the submission with respect to literature.
- One of the advantages mentioned is that CSC-CTRL has reduced computational cost over prior models (line 48 in the paper). But wouldn’t the use of FISTA iterations in the encoder’s forward pass add extra overhead making it more expensive than a conventional autoencoder?
[1] Liu, Tianlin, et al. "Learning multiscale convolutional dictionaries for image reconstruction." IEEE Transactions on Computational Imaging 8 (2022): 425-437
[2] Li, Minghan, et al. "Video rain streak removal by multiscale convolutional sparse coding." Proceedings of the IEEE conference on computer vision and pattern recognition. 2018.
rating: 6
confidence: 3 |
1AEJSYO6GX | Closed-Loop Transcription via Convolutional Sparse Coding | [
"Xili Dai",
"Ke Chen",
"Shengbang Tong",
"Jingyuan Zhang",
"Xingjian Gao",
"Mingyang Li",
"Druv Pai",
"Yuexiang Zhai",
"Xiaojun Yuan",
"Heung-Yeung Shum",
"Lionel Ni",
"Yi Ma"
] | Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets. | [
"Convolutional Sparse Coding",
"Inverse Problem",
"Closed-Loop Transcription"
] | https://openreview.net/pdf?id=1AEJSYO6GX | 8MUago8Fmz | decision | 1,700,363,554,204 | 1AEJSYO6GX | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Program_Chairs"
] | decision: Accept (Oral)
comment: The paper presents the CSC-CTRL framework, which combines convolutional sparse coding (CSC) with the CTRL framework for image generation. Reviewers generally find the paper's contributions promising, with good image generation quality and interpretability in latent codes. However, they raise several concerns and suggestions for improvement, including the need for comparisons with related works that use multi-scale CSC for image generation, clarification on computational costs, and the inclusion of higher-resolution experiments. Reviewers also request additional information such as training times, reconstruction quality metrics, and clearer figures. Overall, the paper has potential but requires some refinements and clarifications to address these concerns.
The action PC chair for this paper is Atlas Wang, who made the decision after carefully reading the paper as well as the comments by all reviewers and AC. The decision is agreed by all PC chairs.
title: Paper Decision |
1AEJSYO6GX | Closed-Loop Transcription via Convolutional Sparse Coding | [
"Xili Dai",
"Ke Chen",
"Shengbang Tong",
"Jingyuan Zhang",
"Xingjian Gao",
"Mingyang Li",
"Druv Pai",
"Yuexiang Zhai",
"Xiaojun Yuan",
"Heung-Yeung Shum",
"Lionel Ni",
"Yi Ma"
] | Autoencoding has achieved great empirical success as a framework for learning generative models for natural images. Autoencoders often use generic deep networks as the encoder or decoder, which are difficult to interpret, and the learned representations lack clear structure. In this work, we make the explicit assumption that the image distribution is generated from a multi-stage sparse deconvolution. The corresponding inverse map, which we use as an encoder, is a multi-stage convolution sparse coding (CSC), with each stage obtained from unrolling an optimization algorithm for solving the corresponding (convexified) sparse coding program. To avoid computational difficulties in minimizing distributional distance between the real and generated images, we utilize the recent closed-loop transcription (CTRL) framework that optimizes the rate reduction of the learned sparse representations. Conceptually, our method has high-level connections to score-matching methods such as diffusion models. Empirically, our framework demonstrates competitive performance on large-scale datasets, such as ImageNet-1K, compared to existing autoencoding and generative methods under fair conditions. Even with simpler networks and fewer computational resources, our method demonstrates high visual quality in regenerated images. More surprisingly, the learned autoencoder performs well on unseen datasets. Our method enjoys several side benefits, including more structured and interpretable representations, more stable convergence, and scalability to large datasets. Our method is arguably the first to demonstrate that a concatenation of multiple convolution sparse coding/decoding layers leads to an interpretable and effective autoencoder for modeling the distribution of large-scale natural image datasets. | [
"Convolutional Sparse Coding",
"Inverse Problem",
"Closed-Loop Transcription"
] | https://openreview.net/pdf?id=1AEJSYO6GX | 2Qf0ZULJ0H | meta_review | 1,699,955,675,410 | 1AEJSYO6GX | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission40/Area_Chair_RcdN"
] | metareview: The CSC-CTRL framework proposed by the authors is an interesting combination of convolutional sparse coding and a distributional loss function related to compression. The resulting autoencoders are more lightweight than the compared convolutional networks, while yielding perceptually good quality images and outperforming some of the comparable generative techniques proposed earlier. Overall it is intriguing to see that this convolutional framework can perform so well and yield strong sample-level encoding.
For balance, I also want to state some places where the manuscript could be improved. The "interpretability" aspect is in my opinion overstated. I don't think that the sparse feature representations are any more or less interpretable than feature maps in convnets, nor is the related generative process more interpretable than a convnet decoder. Since 2 iterations of FISTA are generally not enough to produce the _sparsest_ code, the feature maps here are _some_ sparse feature maps which are optimized end-to-end for generative performance.
Some of the arguments in responses seem to sidestep important issues—for example, why it is not necessary to compare with SOTA. Naming "engineering tricks" as culprits for performance differences sounds somewhat evasive, especially without a detailed elaboration, and acknowledging that the authors work with tiny images. Batch normalization and ReLUs in between the "principled" (de)convolution layers could similarly be declared engineering tricks (without a deeper discussion, at least). Finally, I doubt that CSC strategies can scale to high resolutions without explicitly accounting for regularity across scales and directions, which necessarily involves some kind of multiscale design. But it would be intriguing to be proven wrong.
I agree with the reviewers that this paper is marginally above the acceptance threshold and urge the authors to incorporate the promised changes, including improved comparisons with prior art, in the camera ready version.
recommendation: Accept (Poster)
confidence: 5: The area chair is absolutely certain |
0VU6Vlh0zy | Less is More – Towards parsimonious multi-task models using structured sparsity | [
"Richa Upadhyay",
"Ronald Phlypo",
"Rajkumar Saini",
"Marcus Liwicki"
] | Model sparsification in deep learning promotes simpler, more interpretable models with fewer parameters.
This not only reduces the model's memory footprint and computational needs but also shortens inference time.
This work focuses on creating sparse models optimized for multiple tasks with fewer parameters.
These parsimonious models also possess the potential to match or outperform dense models in terms of performance.
In this work, we introduce channel-wise $l_1/l_2$ group sparsity in the shared convolutional layers parameters (or weights) of the multi-task learning model.
This approach facilitates the removal of extraneous groups i.e., channels (due to $l_1$ regularization) and also imposes a penalty on the weights, further enhancing the learning efficiency for all tasks (due to $l_2$ regularization).
We analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used multi-task learning datasets: NYU-v2 and CelebAMask-HQ.
On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70\% sparsity outperform their dense equivalents.
We also investigate how changing the degree of sparsification influences the model's performance, the overall sparsity percentage, the patterns of sparsity, and the inference time. | [
"Multi-task learning",
"structured sparsity",
"group sparsity",
"parameter pruning",
"semantic segmentation",
"depth estimation",
"surface normal estimation"
] | https://openreview.net/pdf?id=0VU6Vlh0zy | we1bzhtZRt | official_review | 1,697,403,066,662 | 0VU6Vlh0zy | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission26/Reviewer_hye9"
] | title: Official Review for Paper 26
review: The paper proposes to learn structured (group) sparsity in MTL, i.e., it learns sparse shared features among multiple tasks. They analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used Multi-Task Learning (MTL) datasets: NYU-v2 and CelebAMask HQ. On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70% sparsity outperform their dense equivalents. I have following concerns:
1. The paper doesn't clearly explain how their approach bring benefit over many other sparse-learning works like ( https://arxiv.org/pdf/1911.05034.pdf https://arxiv.org/pdf/1705.04886.pdf). The novelty of the work is limited.
2. In addition, the motivation of the work in the introduction is not strongly enlisted (eg. organization of parameters in a CNN, CNNs can develop redundant filter). It will be important to explain why/how channel-wise l1/l2 penalty to the shared (CNN) layer parameters can help in solve complex computer vision tasks (contribution 1).
3. The paper lacks any comparable sparsity-induced MTL baselines in evaluation which is required to show the effectiveness. I am also curious to know how the authors calculated mean inference time in figure 4.
4. How do the authors regulate between Loss1, Loss2, ... Loss N of the proposed architecture in Figure 2. Seems very difficult to tune.
rating: 6
confidence: 3 |
0VU6Vlh0zy | Less is More – Towards parsimonious multi-task models using structured sparsity | [
"Richa Upadhyay",
"Ronald Phlypo",
"Rajkumar Saini",
"Marcus Liwicki"
] | Model sparsification in deep learning promotes simpler, more interpretable models with fewer parameters.
This not only reduces the model's memory footprint and computational needs but also shortens inference time.
This work focuses on creating sparse models optimized for multiple tasks with fewer parameters.
These parsimonious models also possess the potential to match or outperform dense models in terms of performance.
In this work, we introduce channel-wise $l_1/l_2$ group sparsity in the shared convolutional layers parameters (or weights) of the multi-task learning model.
This approach facilitates the removal of extraneous groups i.e., channels (due to $l_1$ regularization) and also imposes a penalty on the weights, further enhancing the learning efficiency for all tasks (due to $l_2$ regularization).
We analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used multi-task learning datasets: NYU-v2 and CelebAMask-HQ.
On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70\% sparsity outperform their dense equivalents.
We also investigate how changing the degree of sparsification influences the model's performance, the overall sparsity percentage, the patterns of sparsity, and the inference time. | [
"Multi-task learning",
"structured sparsity",
"group sparsity",
"parameter pruning",
"semantic segmentation",
"depth estimation",
"surface normal estimation"
] | https://openreview.net/pdf?id=0VU6Vlh0zy | s6nV8VElTk | official_review | 1,696,680,505,651 | 0VU6Vlh0zy | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission26/Reviewer_sZcf"
] | title: Official Review
review: This paper adopts the group lasso penalty for multi-task learning, resulting in cross-task structural sparsity.
The experimental results show that removing redundant parameters further improves MTL performance.
Moreover, this paper shows that appreciating sparsity not only reduces computation costs for each task but also benefits better task performance.
The proposed method is not novel but is effective in learning structured sparsity for multi-task learning, which is potentially utilized in various scenarios.
Considering this paper focuses on the sparsification of hard-shared parameters, the experiments are reasonable, but proper comparison is also helpful.
Like the MTL performance under other structure sparsity methods like SWP[1] and SSL[2].
[1] Pruning Filter in Filter
[2] Learning structured sparsity in deep neural networks
rating: 6: Marginally above acceptance threshold
confidence: 5: The reviewer is absolutely certain that the evaluation is correct and very familiar with the relevant literature |
0VU6Vlh0zy | Less is More – Towards parsimonious multi-task models using structured sparsity | [
"Richa Upadhyay",
"Ronald Phlypo",
"Rajkumar Saini",
"Marcus Liwicki"
] | Model sparsification in deep learning promotes simpler, more interpretable models with fewer parameters.
This not only reduces the model's memory footprint and computational needs but also shortens inference time.
This work focuses on creating sparse models optimized for multiple tasks with fewer parameters.
These parsimonious models also possess the potential to match or outperform dense models in terms of performance.
In this work, we introduce channel-wise $l_1/l_2$ group sparsity in the shared convolutional layers parameters (or weights) of the multi-task learning model.
This approach facilitates the removal of extraneous groups i.e., channels (due to $l_1$ regularization) and also imposes a penalty on the weights, further enhancing the learning efficiency for all tasks (due to $l_2$ regularization).
We analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used multi-task learning datasets: NYU-v2 and CelebAMask-HQ.
On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70\% sparsity outperform their dense equivalents.
We also investigate how changing the degree of sparsification influences the model's performance, the overall sparsity percentage, the patterns of sparsity, and the inference time. | [
"Multi-task learning",
"structured sparsity",
"group sparsity",
"parameter pruning",
"semantic segmentation",
"depth estimation",
"surface normal estimation"
] | https://openreview.net/pdf?id=0VU6Vlh0zy | olTViOnvOK | meta_review | 1,699,761,314,258 | 0VU6Vlh0zy | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission26/Area_Chair_aqca"
] | metareview: This paper proposes an approach that incorporates group sparsity into the shared CNN backbone in multi-task learning. The proposed approach is extensively evaluated on two datasets with heterogeneous CV tasks, showing that it can significantly prune the model parameters while having improved performance in comparison to non-sparse baselines.
During the rebuttal, the authors successfully clarified reviewers' questions about various technical details. There is still one major concern raised by all reviewers about the lack of comparison to various existing sparsity methods, which undermines the novelty of the proposed approach. However, reviewers agreed that the merits of this paper slightly overweigh its shortcoming.
recommendation: Accept (Poster)
confidence: 4: The area chair is confident but not absolutely certain |
0VU6Vlh0zy | Less is More – Towards parsimonious multi-task models using structured sparsity | [
"Richa Upadhyay",
"Ronald Phlypo",
"Rajkumar Saini",
"Marcus Liwicki"
] | Model sparsification in deep learning promotes simpler, more interpretable models with fewer parameters.
This not only reduces the model's memory footprint and computational needs but also shortens inference time.
This work focuses on creating sparse models optimized for multiple tasks with fewer parameters.
These parsimonious models also possess the potential to match or outperform dense models in terms of performance.
In this work, we introduce channel-wise $l_1/l_2$ group sparsity in the shared convolutional layers parameters (or weights) of the multi-task learning model.
This approach facilitates the removal of extraneous groups i.e., channels (due to $l_1$ regularization) and also imposes a penalty on the weights, further enhancing the learning efficiency for all tasks (due to $l_2$ regularization).
We analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used multi-task learning datasets: NYU-v2 and CelebAMask-HQ.
On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70\% sparsity outperform their dense equivalents.
We also investigate how changing the degree of sparsification influences the model's performance, the overall sparsity percentage, the patterns of sparsity, and the inference time. | [
"Multi-task learning",
"structured sparsity",
"group sparsity",
"parameter pruning",
"semantic segmentation",
"depth estimation",
"surface normal estimation"
] | https://openreview.net/pdf?id=0VU6Vlh0zy | FTRoQBNLTi | decision | 1,700,497,595,521 | 0VU6Vlh0zy | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Program_Chairs"
] | decision: Accept (Oral)
comment: The authors propose a method for learning structured sparsity in multi-task learning. Overall, the reviewers and AC agree that the approach proposed in the paper is an interesting contribution despite some missing comparisons and ablations. The method is evaluated on two widely-used multi-task learning datasets, showing that it can significantly prune the model parameters while having improved performance in comparison to non-sparse baselines. For the camera-ready version, the authors should strongly consider providing a more detailed explanation and experiments of how the proposed approach brings benefit over many other structured sparse-learning approaches.
The action PC chair for this paper is Gintare Karolina Dziugaite, who made the decision after carefully reading the paper as well as the comments by all reviewers and AC. The decision is agreed by all PC chairs.
title: Paper Decision |
0VU6Vlh0zy | Less is More – Towards parsimonious multi-task models using structured sparsity | [
"Richa Upadhyay",
"Ronald Phlypo",
"Rajkumar Saini",
"Marcus Liwicki"
] | Model sparsification in deep learning promotes simpler, more interpretable models with fewer parameters.
This not only reduces the model's memory footprint and computational needs but also shortens inference time.
This work focuses on creating sparse models optimized for multiple tasks with fewer parameters.
These parsimonious models also possess the potential to match or outperform dense models in terms of performance.
In this work, we introduce channel-wise $l_1/l_2$ group sparsity in the shared convolutional layers parameters (or weights) of the multi-task learning model.
This approach facilitates the removal of extraneous groups i.e., channels (due to $l_1$ regularization) and also imposes a penalty on the weights, further enhancing the learning efficiency for all tasks (due to $l_2$ regularization).
We analyzed the results of group sparsity in both single-task and multi-task settings on two widely-used multi-task learning datasets: NYU-v2 and CelebAMask-HQ.
On both datasets, which consist of three different computer vision tasks each, multi-task models with approximately 70\% sparsity outperform their dense equivalents.
We also investigate how changing the degree of sparsification influences the model's performance, the overall sparsity percentage, the patterns of sparsity, and the inference time. | [
"Multi-task learning",
"structured sparsity",
"group sparsity",
"parameter pruning",
"semantic segmentation",
"depth estimation",
"surface normal estimation"
] | https://openreview.net/pdf?id=0VU6Vlh0zy | DFPWCtduzU | official_review | 1,696,686,030,493 | 0VU6Vlh0zy | [
"everyone"
] | [
"CPAL.cc/2024/Conference/Submission26/Reviewer_mbZZ"
] | title: Official Review of Paper26 by Reviewer mbZZ
review: The author introduces channel-wise l1/l2 group sparsity into the shared convolutional layer parameters (or weights) of multi-task learning models. However, lack of comparative analysis with other relevant sparse multitask learning methods limits its persuasiveness.
Pros:
+ Extensive experiments.
Cons:
+ Some paragraphs present unclear logic, such as: " .. Therefore, Sparsity aids in achieving parsimony ..." does not have a clear causal relationship with its preceding context.
+ Lacks comparative analysis with any other methods, such as [1]
[1] Sun T, Shao Y, Li X, et al. Learning sparse sharing architectures for multiple tasks[C]//Proceedings of the AAAI conference on artificial intelligence. 2020, 34(05): 8936-8943.
rating: 6
confidence: 4 |
ogXYJc1iE9 | The effect of age on the immediate recall of source vs. goal information from Czech sentences | [
"Eva Pospíšilová"
] | Background: Earlier studies have observed systematic differences in speakers' ability to remember information from sentences [1]. Research focusing on the English language has shown that English speakers tend to recall the actor and subject of a sentence more easily, whereas information conveyed through verbs, adjectives, or adverbs is more difficult to remember [2]. These systematic differences have also been observed among Czech speakers; people tend to remember information conveyed by the object better than information provided by attributes or adverbial adjuncts specifying place or time [3, 4]. However, there is currently a gap in both Czech and international literature regarding research on sentence recall in relation to developmental aspects. While tasks focusing on sentence processing in adults have observed so-called good-enough processing mechanisms [5], reliance on good-enough processing has been less frequently shown in children and adolescents [6]. Therefore, it can be expected that the ability to memorize and recall information from a sentence will differ between children and adults and that these abilities will change during cognitive development. The present study thus investigates the effect of age on the ability to immediately recall information from sentences in children aged 11 to 17 and adults, focusing on the recall of information from adverbial adjuncts specifying direction/goal and source/place of origin. The selection of these specific types of adjuncts was based on findings from previous studies that pointed to a bias for goal information in language processing [7].
Method: Using the online dictionaries Vallex 4.5 [8] and SynSem Class Lexicon 5.0 [9], a total of 24 Czech transitive verbs were selected that can be combined with adverbial adjuncts conveying both information about the source and the direction/goal of the action. For each verb, a set of 4 sentences was created, two of which contain the verb in combination with an adverbial adjunct conveying information about direction, while the other two feature the verb with information about the source. The placement of these adverbial adjuncts also differed across the 4 sentences (see Table 1). In total, the experiment will contain 24 experimental and 72 filler sentences, which are currently being created. The experiment will use a self-paced reading method, each sentence will be immediately followed by an open-ended comprehension question targeting the information from adverbial adjunct conveying goal/source information. The type of sentence and question from each experimental item will be assigned to participants according to a Latin square design. The experiment will be tested on a group of children aged 11 to 17 years and on a control group of adults over 20 years of age.
Expected results: Consistent with previous studies, we expect to observe systematic differences in sentence recall among adult speakers, specifically that Czech adults will recall goal information better than source information. For the youngest group of child participants, we expect no such differences, as they likely have not yet acquired mechanisms of the good-enough language processing. Furthermore, we anticipate that the degree of systematic differences in recalling information from the two selected adverbial adjuncts will be modulated by age, with more systematic differences emerging as participants get older.
Key words: information recall, self-paced reading, source, goal, developmental aspects | [
"information recall",
"self-paced reading",
"source",
"goal",
"developmental aspects"
] | https://openreview.net/pdf?id=ogXYJc1iE9 | tvWvNvJPKQ | official_review | 1,736,340,787,182 | ogXYJc1iE9 | [
"everyone"
] | [
"~Radek_Šimík1"
] | title: Very nice and clearly structured abstract!
review: I have some contents-related comments and a few minor ones.
Comments on substance: It's a bit unclear to me whether there's little evidence for good-enough processing in children because so few people have tried so far, or because people have failed to find it; in other words, it's unclear whether [6] shows evidence for or against it. Normally it's fine to remain a bit ambivalent in these cases, but this is very central to your overall aims.
More out of curiosity: is good-enough processing something that has to be learned? Isn't it a very general mechanism that one starts with (and potentially needs to avoid it if necessary - when one needs to concentrate hard). I'd appreciate a bit more info or references on this issue.
- A suggestion for shortening: "Research focusing on the English language has that" could be dropped completely.
- Since the sentence that follows says basically the same for Czech, the two claims can be connected. E.g. "It has been shown that speakers... (for English: [2]; for Czech: [3, 4])."
- "in both Czech and international literature" could be dropped (generally, I'd advise you to avoid this opposition altogether, it's a bad habit of Czech linguists to speak about "Czech" vs. "international"... it's all just science)
- I'd add an "and" between "self-paced reading method" and "each..."
- "a Latin square design" - I guess "the Latin square design" makes more sense (it's a unique concept) |
ogXYJc1iE9 | The effect of age on the immediate recall of source vs. goal information from Czech sentences | [
"Eva Pospíšilová"
] | Background: Earlier studies have observed systematic differences in speakers' ability to remember information from sentences [1]. Research focusing on the English language has shown that English speakers tend to recall the actor and subject of a sentence more easily, whereas information conveyed through verbs, adjectives, or adverbs is more difficult to remember [2]. These systematic differences have also been observed among Czech speakers; people tend to remember information conveyed by the object better than information provided by attributes or adverbial adjuncts specifying place or time [3, 4]. However, there is currently a gap in both Czech and international literature regarding research on sentence recall in relation to developmental aspects. While tasks focusing on sentence processing in adults have observed so-called good-enough processing mechanisms [5], reliance on good-enough processing has been less frequently shown in children and adolescents [6]. Therefore, it can be expected that the ability to memorize and recall information from a sentence will differ between children and adults and that these abilities will change during cognitive development. The present study thus investigates the effect of age on the ability to immediately recall information from sentences in children aged 11 to 17 and adults, focusing on the recall of information from adverbial adjuncts specifying direction/goal and source/place of origin. The selection of these specific types of adjuncts was based on findings from previous studies that pointed to a bias for goal information in language processing [7].
Method: Using the online dictionaries Vallex 4.5 [8] and SynSem Class Lexicon 5.0 [9], a total of 24 Czech transitive verbs were selected that can be combined with adverbial adjuncts conveying both information about the source and the direction/goal of the action. For each verb, a set of 4 sentences was created, two of which contain the verb in combination with an adverbial adjunct conveying information about direction, while the other two feature the verb with information about the source. The placement of these adverbial adjuncts also differed across the 4 sentences (see Table 1). In total, the experiment will contain 24 experimental and 72 filler sentences, which are currently being created. The experiment will use a self-paced reading method, each sentence will be immediately followed by an open-ended comprehension question targeting the information from adverbial adjunct conveying goal/source information. The type of sentence and question from each experimental item will be assigned to participants according to a Latin square design. The experiment will be tested on a group of children aged 11 to 17 years and on a control group of adults over 20 years of age.
Expected results: Consistent with previous studies, we expect to observe systematic differences in sentence recall among adult speakers, specifically that Czech adults will recall goal information better than source information. For the youngest group of child participants, we expect no such differences, as they likely have not yet acquired mechanisms of the good-enough language processing. Furthermore, we anticipate that the degree of systematic differences in recalling information from the two selected adverbial adjuncts will be modulated by age, with more systematic differences emerging as participants get older.
Key words: information recall, self-paced reading, source, goal, developmental aspects | [
"information recall",
"self-paced reading",
"source",
"goal",
"developmental aspects"
] | https://openreview.net/pdf?id=ogXYJc1iE9 | 8NictiZMKo | official_review | 1,736,258,537,442 | ogXYJc1iE9 | [
"everyone"
] | [
"~Maria_Onoeva1"
] | title: Good job!
review: The abstract outlines a planned self-paced reading study that will investigate how age affects the ability to immediately recall information from sentences. It’s well-written and structured, and the text is easy to follow. The research question is clear, but I have a few comments about the design of the experiment (see below). Including an example item really helped clarify the study design, but you’ll need to translate it if you’re planning to submit the abstract to an international conference. One more thing, I’m a bit confused by the terminology. Is the distinction between direction/goal and source/place of origin? In the example table and the text, you seem to use both, so maybe it would make sense to stick to one from the two.
1) What exactly do you mean by "an open-ended comprehension question"? Are participants supposed to type their answers? What is going to be measured as the dependent variable?
2) Why is this a self-paced reading study? Are you measuring reading time too? If so, do you have any predictions about age, reading speed and recall?
3) Since you’re manipulating the position of the adjuncts, do you have any predictions about how that might affect the results?
Good luck with the study! |
ogXYJc1iE9 | The effect of age on the immediate recall of source vs. goal information from Czech sentences | [
"Eva Pospíšilová"
] | Background: Earlier studies have observed systematic differences in speakers' ability to remember information from sentences [1]. Research focusing on the English language has shown that English speakers tend to recall the actor and subject of a sentence more easily, whereas information conveyed through verbs, adjectives, or adverbs is more difficult to remember [2]. These systematic differences have also been observed among Czech speakers; people tend to remember information conveyed by the object better than information provided by attributes or adverbial adjuncts specifying place or time [3, 4]. However, there is currently a gap in both Czech and international literature regarding research on sentence recall in relation to developmental aspects. While tasks focusing on sentence processing in adults have observed so-called good-enough processing mechanisms [5], reliance on good-enough processing has been less frequently shown in children and adolescents [6]. Therefore, it can be expected that the ability to memorize and recall information from a sentence will differ between children and adults and that these abilities will change during cognitive development. The present study thus investigates the effect of age on the ability to immediately recall information from sentences in children aged 11 to 17 and adults, focusing on the recall of information from adverbial adjuncts specifying direction/goal and source/place of origin. The selection of these specific types of adjuncts was based on findings from previous studies that pointed to a bias for goal information in language processing [7].
Method: Using the online dictionaries Vallex 4.5 [8] and SynSem Class Lexicon 5.0 [9], a total of 24 Czech transitive verbs were selected that can be combined with adverbial adjuncts conveying both information about the source and the direction/goal of the action. For each verb, a set of 4 sentences was created, two of which contain the verb in combination with an adverbial adjunct conveying information about direction, while the other two feature the verb with information about the source. The placement of these adverbial adjuncts also differed across the 4 sentences (see Table 1). In total, the experiment will contain 24 experimental and 72 filler sentences, which are currently being created. The experiment will use a self-paced reading method, each sentence will be immediately followed by an open-ended comprehension question targeting the information from adverbial adjunct conveying goal/source information. The type of sentence and question from each experimental item will be assigned to participants according to a Latin square design. The experiment will be tested on a group of children aged 11 to 17 years and on a control group of adults over 20 years of age.
Expected results: Consistent with previous studies, we expect to observe systematic differences in sentence recall among adult speakers, specifically that Czech adults will recall goal information better than source information. For the youngest group of child participants, we expect no such differences, as they likely have not yet acquired mechanisms of the good-enough language processing. Furthermore, we anticipate that the degree of systematic differences in recalling information from the two selected adverbial adjuncts will be modulated by age, with more systematic differences emerging as participants get older.
Key words: information recall, self-paced reading, source, goal, developmental aspects | [
"information recall",
"self-paced reading",
"source",
"goal",
"developmental aspects"
] | https://openreview.net/pdf?id=ogXYJc1iE9 | 1Pv1mgp9Cb | official_review | 1,736,277,973,914 | ogXYJc1iE9 | [
"everyone"
] | [
"~Anna_Staňková1"
] | title: Great!
review: This is a well-written abstract about a planned study. It has a clear structure and itt is easy to follow. I like how you use numbers in brackets for citations - it saves space, yet you get the important references in the text. I also appreciate that you hyperlinked Vallex and SynSemClass Lexicon since it makes access a lot easier. In the method part, I would suggest describing the item as a 2 x 2 design (2 variables with two values), not as "a set of 4 sentences". |
Vs3myOAvgk | Discrimination of Native Vowels in Bilingual and Monolingual Czech Infants with Familial Risk of Dyslexia | [
"Kateřina Kynčlová"
] | Developmental dyslexia is a neurodevelopmental disorder linked to phonological and speech processing deficits, with a strong hereditary component that allows for the allocation of at-risk children with a higher susceptibility to the disorder (Kalashnikova et al., 2020). This study investigates whether Czech at-risk infants show a reduced ability to discriminate native vowels and whether bilingualism modulates the early speech sound development. The study focuses on two types of vowel contrasts that yield different discrimination patterns in typically developing Czech infants–contrasts cued by spectral properties vs. contrasts cued by duration. We hypothesise that at-risk infants will show weaker vowel discrimination, particularly for length contrasts, and that bilingual at-risk infants will exhibit smaller discrimination deficits than monolingual at-risk infants, potentially due to enhanced cognitive and perceptual skills.
The study aims to test 96 Czech infants aged 3.5-7.5M, divided into monolingual (n=48; 24 at-risk) and bilingual (n=48; 24 at-risk) groups. Infants will be tested in a central fixation paradigm. They will be familiarised with strings of naturally produced Czech syllables [fa], and after reaching a habituation criterion, their discrimination of Czech phonemic [fa]-[fe] and [fa]-[fa:] contrasts will be tested with alternating vs. non-alternating trials.
We predict impaired discrimination of vowel length in at-risk infants, consistent with studies showing deficits in consonant length discrimination in at-risk children in languages with phonemic consonantal length (Gerrits et al., 2008; Richardson et al., 2003). Alternatively, at-risk infants acquiring Czech may not be disadvantaged in vowel length discrimination, as this ability seems to be developed in typically developing Czech infants from a very early age (Chládková et al., 2021; Paillereau et al., 2021). Literature on early speech perception development in at-risk bilinguals is lacking and the present results will thus profoundly broaden our understanding of early speech development in bilingual at-risk infants. | [
"Speech perception",
"vowel discrimination",
"developmental dyslexia",
"at-risk infants",
"early bilingualism"
] | https://openreview.net/pdf?id=Vs3myOAvgk | wKMLSlqP8r | official_review | 1,736,168,814,748 | Vs3myOAvgk | [
"everyone"
] | [
"~Maria_Onoeva1"
] | title: Good job!
review: This is a nice and clear abstract for the Workshop on Infant Language Development (WILD) 2025. Given the limited space, it’s well-structured and includes all the necessary background and methodological details. I learned a lot about child speech processing and would love to hear more. Best of luck with the submission and the research, I’m looking forward to the results!
I have the following questions to the research:
1) What are the languages for bilingual infants? Czech + another?
2) Are you going to test these syllables only? Why not to test words where these syllables occur in? I guess this ensures clearer results but in real life people do not produce syllables only.
3) Is it possible to track and test these infants further, e.g., when they are todlers? They are at-risk but it doen't mean they will necessary inherit dyslexia, right? Or maybe it works in a different way, I have to learn more. |
Vs3myOAvgk | Discrimination of Native Vowels in Bilingual and Monolingual Czech Infants with Familial Risk of Dyslexia | [
"Kateřina Kynčlová"
] | Developmental dyslexia is a neurodevelopmental disorder linked to phonological and speech processing deficits, with a strong hereditary component that allows for the allocation of at-risk children with a higher susceptibility to the disorder (Kalashnikova et al., 2020). This study investigates whether Czech at-risk infants show a reduced ability to discriminate native vowels and whether bilingualism modulates the early speech sound development. The study focuses on two types of vowel contrasts that yield different discrimination patterns in typically developing Czech infants–contrasts cued by spectral properties vs. contrasts cued by duration. We hypothesise that at-risk infants will show weaker vowel discrimination, particularly for length contrasts, and that bilingual at-risk infants will exhibit smaller discrimination deficits than monolingual at-risk infants, potentially due to enhanced cognitive and perceptual skills.
The study aims to test 96 Czech infants aged 3.5-7.5M, divided into monolingual (n=48; 24 at-risk) and bilingual (n=48; 24 at-risk) groups. Infants will be tested in a central fixation paradigm. They will be familiarised with strings of naturally produced Czech syllables [fa], and after reaching a habituation criterion, their discrimination of Czech phonemic [fa]-[fe] and [fa]-[fa:] contrasts will be tested with alternating vs. non-alternating trials.
We predict impaired discrimination of vowel length in at-risk infants, consistent with studies showing deficits in consonant length discrimination in at-risk children in languages with phonemic consonantal length (Gerrits et al., 2008; Richardson et al., 2003). Alternatively, at-risk infants acquiring Czech may not be disadvantaged in vowel length discrimination, as this ability seems to be developed in typically developing Czech infants from a very early age (Chládková et al., 2021; Paillereau et al., 2021). Literature on early speech perception development in at-risk bilinguals is lacking and the present results will thus profoundly broaden our understanding of early speech development in bilingual at-risk infants. | [
"Speech perception",
"vowel discrimination",
"developmental dyslexia",
"at-risk infants",
"early bilingualism"
] | https://openreview.net/pdf?id=Vs3myOAvgk | mogRtV4qWg | official_review | 1,736,278,633,926 | Vs3myOAvgk | [
"everyone"
] | [
"~Anna_Staňková1"
] | title: Great!
review: This abstract is well-written and well-structured. I especially like how it is simple to understand even for someone who is not an expert on the topic. Minor comment: I would suggest placing the very last sentence ("Literature on early speech perception ... in biligual at-risk infants") to the first paragraph, maybe in a reformulated way. To me it sounds as an important reason why you are doing this research, so I would stress it by moving it rather to the beginning of the abstract. |
Vs3myOAvgk | Discrimination of Native Vowels in Bilingual and Monolingual Czech Infants with Familial Risk of Dyslexia | [
"Kateřina Kynčlová"
] | Developmental dyslexia is a neurodevelopmental disorder linked to phonological and speech processing deficits, with a strong hereditary component that allows for the allocation of at-risk children with a higher susceptibility to the disorder (Kalashnikova et al., 2020). This study investigates whether Czech at-risk infants show a reduced ability to discriminate native vowels and whether bilingualism modulates the early speech sound development. The study focuses on two types of vowel contrasts that yield different discrimination patterns in typically developing Czech infants–contrasts cued by spectral properties vs. contrasts cued by duration. We hypothesise that at-risk infants will show weaker vowel discrimination, particularly for length contrasts, and that bilingual at-risk infants will exhibit smaller discrimination deficits than monolingual at-risk infants, potentially due to enhanced cognitive and perceptual skills.
The study aims to test 96 Czech infants aged 3.5-7.5M, divided into monolingual (n=48; 24 at-risk) and bilingual (n=48; 24 at-risk) groups. Infants will be tested in a central fixation paradigm. They will be familiarised with strings of naturally produced Czech syllables [fa], and after reaching a habituation criterion, their discrimination of Czech phonemic [fa]-[fe] and [fa]-[fa:] contrasts will be tested with alternating vs. non-alternating trials.
We predict impaired discrimination of vowel length in at-risk infants, consistent with studies showing deficits in consonant length discrimination in at-risk children in languages with phonemic consonantal length (Gerrits et al., 2008; Richardson et al., 2003). Alternatively, at-risk infants acquiring Czech may not be disadvantaged in vowel length discrimination, as this ability seems to be developed in typically developing Czech infants from a very early age (Chládková et al., 2021; Paillereau et al., 2021). Literature on early speech perception development in at-risk bilinguals is lacking and the present results will thus profoundly broaden our understanding of early speech development in bilingual at-risk infants. | [
"Speech perception",
"vowel discrimination",
"developmental dyslexia",
"at-risk infants",
"early bilingualism"
] | https://openreview.net/pdf?id=Vs3myOAvgk | btyRS7Mawg | official_review | 1,736,175,188,986 | Vs3myOAvgk | [
"everyone"
] | [
"~Lucie_Jarůšková1"
] | title: Great abstract
review: The psycholinguistic topic of the paper (*dyslexia and early language perception*) is current, interesting and socially relevant. I especially appreciate that the author does not only examine comparisons between at-risk and typically developing monolingual children, but also includes the bilingual population.
The abstract is written in clear, professional, and excellent English.
In terms of structure, the author clearly defines the issue in the introduction and moves smoothly into a description of the method. If the space of the text allowed and the author had it planned, it would have been useful to give more details about the experiment itself, for example how long it takes, or how many *trials* (items, stimuli) it contains in total – a few words would have sufficed. Since this is not a finished experiment with results, but a design proposal, this section is essential.
Hypotheses/predictions are listed in two places, they are a bit repetitive, but I understand the repetition in the part of conclusion, as there is an explanation based on literature. It would be appropriate to add to the second hypothesis whether the differences are supposed to be in one of the contrasts (length or spectral quality) or both contrasts – by adding e.g. *in both contrasts*.
My question:
1. Will the author also compare non-at-risk infants, typically developing monolingual and bilingual?
Overall, this is an excellent and readable abstract that summarizes and presents an interesting experiment, suitable for the conference. Fingers crossed with acceptance! |
Vs3myOAvgk | Discrimination of Native Vowels in Bilingual and Monolingual Czech Infants with Familial Risk of Dyslexia | [
"Kateřina Kynčlová"
] | Developmental dyslexia is a neurodevelopmental disorder linked to phonological and speech processing deficits, with a strong hereditary component that allows for the allocation of at-risk children with a higher susceptibility to the disorder (Kalashnikova et al., 2020). This study investigates whether Czech at-risk infants show a reduced ability to discriminate native vowels and whether bilingualism modulates the early speech sound development. The study focuses on two types of vowel contrasts that yield different discrimination patterns in typically developing Czech infants–contrasts cued by spectral properties vs. contrasts cued by duration. We hypothesise that at-risk infants will show weaker vowel discrimination, particularly for length contrasts, and that bilingual at-risk infants will exhibit smaller discrimination deficits than monolingual at-risk infants, potentially due to enhanced cognitive and perceptual skills.
The study aims to test 96 Czech infants aged 3.5-7.5M, divided into monolingual (n=48; 24 at-risk) and bilingual (n=48; 24 at-risk) groups. Infants will be tested in a central fixation paradigm. They will be familiarised with strings of naturally produced Czech syllables [fa], and after reaching a habituation criterion, their discrimination of Czech phonemic [fa]-[fe] and [fa]-[fa:] contrasts will be tested with alternating vs. non-alternating trials.
We predict impaired discrimination of vowel length in at-risk infants, consistent with studies showing deficits in consonant length discrimination in at-risk children in languages with phonemic consonantal length (Gerrits et al., 2008; Richardson et al., 2003). Alternatively, at-risk infants acquiring Czech may not be disadvantaged in vowel length discrimination, as this ability seems to be developed in typically developing Czech infants from a very early age (Chládková et al., 2021; Paillereau et al., 2021). Literature on early speech perception development in at-risk bilinguals is lacking and the present results will thus profoundly broaden our understanding of early speech development in bilingual at-risk infants. | [
"Speech perception",
"vowel discrimination",
"developmental dyslexia",
"at-risk infants",
"early bilingualism"
] | https://openreview.net/pdf?id=Vs3myOAvgk | PKRiCwzpeD | official_review | 1,736,338,720,894 | Vs3myOAvgk | [
"everyone"
] | [
"~Radek_Šimík1"
] | title: Excellent abstract!
review: This is an excellent abstract. Each sentence has its clear function in the structure of abstract, in line with the standard structure. The abstract is sufficiently short and clearly structured so that section labeling is not necessary. Depending on whether the results are in place before submission, you might consider distributing some of the contents from the last paragraph to previous sections. |
Rb39iWQ6ZR | Processing of Emotional Words by Czech-German Bilinguals | [
"Vojtěch Kocourek"
] | Abstract of my Master's thesis "Processing of Emotional Words by Czech-German Bilinguals". | [
"bilingualism",
"emotionality",
"arousal",
"valence",
"context of learning"
] | https://openreview.net/pdf?id=Rb39iWQ6ZR | wEmVJG5NK6 | official_review | 1,736,346,353,337 | Rb39iWQ6ZR | [
"everyone"
] | [
"~Barbora_Genserová1"
] | title: Great!
review: Clearly structurred abstract written in high-level English. For more space (if needed), the section names could be in bold and not on a separate line and numbered references in text.
I would expect more information about the selection of participants and how (if) they differed in their level of German, and of the words used in the experiment. |
Rb39iWQ6ZR | Processing of Emotional Words by Czech-German Bilinguals | [
"Vojtěch Kocourek"
] | Abstract of my Master's thesis "Processing of Emotional Words by Czech-German Bilinguals". | [
"bilingualism",
"emotionality",
"arousal",
"valence",
"context of learning"
] | https://openreview.net/pdf?id=Rb39iWQ6ZR | Z7qr1OPz3c | official_review | 1,736,336,088,872 | Rb39iWQ6ZR | [
"everyone"
] | [
"~Radek_Šimík1"
] | title: Very nice abstract with clearly stated objectives, method and results
review: I don't have many comments on form. The abstract is clearly and correctly structured and clearly written. Maybe just one minor point: better quality of figures would be appreciated (ideally in vector format, svg, or embedded pdf).
Some comments on substance:
- The participant sample is well-characterized, but we learn little from the abstract about the words that were evaluated. How were they chosen? Did they fall into different categories? If yes, it would have been good to have the different categories visualized separately (e.g. by different colors of datapoints in the plots).
- I'm not sure it's possible to draw the inference from this work that the "acquisition happened in a sufficiently emotional context". That can be inferred only if it is really the case that emotional resonance is the result of "emotional learning". But this is not stated as a fact in the introduction, but rather just a hypothesis. So either the hypothesis si not a hypothesis, but a well-supported fact, or the inference is conditional on the the hypothesis being borne out in the future. |
Rb39iWQ6ZR | Processing of Emotional Words by Czech-German Bilinguals | [
"Vojtěch Kocourek"
] | Abstract of my Master's thesis "Processing of Emotional Words by Czech-German Bilinguals". | [
"bilingualism",
"emotionality",
"arousal",
"valence",
"context of learning"
] | https://openreview.net/pdf?id=Rb39iWQ6ZR | ETxqkJwf4g | official_review | 1,736,166,821,763 | Rb39iWQ6ZR | [
"everyone"
] | [
"~Maria_Onoeva1"
] | title: Good job!
review: A very curious design of the experiment, I'm sorry there was a technical issue, it would be great to look at that type of data. The research questions are clearly stated with respect to the previous work and literature. The results are particularly interesting in relation to the hypotheses, although more research is of course needed. I also appreciate how well it is written and structured, especially given the limited space.
I have the following questions:
1) I didn't quite get what types of words were tested, was it a special category ("emotional words"?) or just any word? How were they picked for the experiment? It is useful to provide a couple of examples or a link to a repository with a table containing average ratings.
2) Another question is about age of participants, were they roughly the same age? Could also gender influence the results?
3) Does it make sense to test Czech speakers for similar words in Czech and then compare their reactions to German words?
Typographical suggestions:
- Bold face section names on one line with paragraphs for more space |
Rb39iWQ6ZR | Processing of Emotional Words by Czech-German Bilinguals | [
"Vojtěch Kocourek"
] | Abstract of my Master's thesis "Processing of Emotional Words by Czech-German Bilinguals". | [
"bilingualism",
"emotionality",
"arousal",
"valence",
"context of learning"
] | https://openreview.net/pdf?id=Rb39iWQ6ZR | 8su7CHP2PK | official_review | 1,736,176,384,362 | Rb39iWQ6ZR | [
"everyone"
] | [
"~Lucie_Jarůšková1"
] | title: Great abstract
review: The topic of the abstract is an interesting one, studied in modern psycholinguistics, and contributes to a better understanding of bilingualism as such and the perception of emotional words.
It is written in good professional English.
The structure of the abstract is clear, partly thanks to the use of headings. I recommend saving space and placing the headings directly next to the text (no line breaks). The abstract presents the subject matter and the three main theoretical points well. There is a slight lack of information about how the words were selected (the preparation of stimuli), although it is understandable that space is limited.
The statistical model used to analyze and compare the groups could be briefly mentioned in the results, which would help to present the results better. This is not an *error*, but a recommendation for improvement and better "selling" of the results.
My questions:
1. It would also have been useful to indicate whether bilingual participants differed in their level of German proficiency (as the abstract already presents the results; it is not automatic that all bilingual speakers reach C2 in all domains). The text shows that the age of acquisition was taken as a factor. Does this mean that their level of German was same or different?
2. And, if / how was the *"the context in which they learned German"* and especially emotional words examined more? Were they dominant in Czech or German?
I appreciate the addition of a graph of the results and a screenshot illustrating the design of the experiment itself. The inclusion of references and mention of the limitations of the study are also great.
Overall, this is a very good abstract that presents an interesting field and experiment. Looking forward to more studies and results! |
Q3n9da9OQ2 | Thinking for Speaking in Context Rather than a Vacuum – A Path Towards Experimental Studies | [
"Tereza Pavlíková"
] | Same as the pdf. | [
"thinking for speaking",
"seeing for speaking",
"aspect",
"conceptualization",
"bilingualism"
] | https://openreview.net/pdf?id=Q3n9da9OQ2 | yBR7RKYsDy | official_review | 1,736,298,680,802 | Q3n9da9OQ2 | [
"everyone"
] | [
"~Matúš_Godál2"
] | title: Review
review: This abstract presents an experiment with the focus on the role of linguistic context in the conceptualization process as described by Slobin. The topic itself seems promising as it addresses the gap in the current research by incorporating contextual linguistic input into the design of experimental studies. The abstract provides the readership with a detailed theoretical background by ancoring it within an already established framework. Although this is a strength of the abstract, clear statement of the research questions or hypotheses is missing. Also, the text of the abstract in its entirety would benefit from its structuring (e.g. IMRaD format), facilitating better cohesion of the text. In addition to that, the provision of more information about the studied population would be appreciated; the readership is only informed about the number of the bilinguals who took part in the experiment. In regard to the presentation of the results, tbe abstract does not mention any statistical results that would strenghten the credibility of the findings even though no significant differences between the monolingual and mixed-language blocks were detected by the author. |
Q3n9da9OQ2 | Thinking for Speaking in Context Rather than a Vacuum – A Path Towards Experimental Studies | [
"Tereza Pavlíková"
] | Same as the pdf. | [
"thinking for speaking",
"seeing for speaking",
"aspect",
"conceptualization",
"bilingualism"
] | https://openreview.net/pdf?id=Q3n9da9OQ2 | QwFbtadJmn | official_review | 1,736,255,632,565 | Q3n9da9OQ2 | [
"everyone"
] | [
"~Maria_Onoeva1"
] | title: Good job!
review: As someone who’s not super familiar with the theoretical foundations of the study, I really appreciated the background you provided earlier. But to be honest, I’m not entirely sure I understood the research questions or hypotheses for this study. I suggest describing them a bit more precisely in the future. Also I noticed that in studies like this, it’s common to include basic demographic details about participants, mention where the study took place and provide statistical results even if they are not significant.
I have a few general questions, and maybe you could point me toward some relevant literature?
1) Why didn’t you test the second incongruent condition, where participants would hear Czech audio but produce English?
2) From what I understood, you only analyzed eye movements, am I right? What about analyzing speech production? I’m not exactly sure how you’d do that, but it seems like it would be crucial for testing linguistic pre-activation.
3) I’m really curious about linguistic pre-activation. Did someone test non-bilingual participants in the same way? I’d be interested to see results from people who don’t speak English (or any other foreign language) but still listen to it. Could that still count as linguistic pre-activation, since it’s a language, even if they don’t speak it? And how would their results compare to people who listened to random, non-linguistic noise? |
Q3n9da9OQ2 | Thinking for Speaking in Context Rather than a Vacuum – A Path Towards Experimental Studies | [
"Tereza Pavlíková"
] | Same as the pdf. | [
"thinking for speaking",
"seeing for speaking",
"aspect",
"conceptualization",
"bilingualism"
] | https://openreview.net/pdf?id=Q3n9da9OQ2 | Blrl2fQTDp | official_review | 1,736,339,840,159 | Q3n9da9OQ2 | [
"everyone"
] | [
"~Radek_Šimík1"
] | title: Very nice abstract with a clear structure, just minor notes below
review: - I really like the the intro - jumping right into the issue.
- I think the adverb "notably" is superfluous. In such a short abstract, everything should be notable. Also, you might want to state already at this point that this is what you are going to do and add that nobody has done it before.
- A minor terminological note: "grammatical aspect" might be a bit misleading (garden-path-y), as it could be understand as "an aspect/issue of grammar". Maybe "verbal aspect" or "the category of aspect" are clearer.
- A related note on contents: Since Czech and English are so different in terms of how they express aspectual properties of events, it remains unclear what is meant by "aspect" here.
- "listening to unrelated background audio" - for some reason I imagined music or something, saying that it's a linguistic audio stimulus would be beneficial.
- "In our first, descriptive analyses" seems superfluous.
- "the effects of context" or rather "the effect of context"? |
Q3n9da9OQ2 | Thinking for Speaking in Context Rather than a Vacuum – A Path Towards Experimental Studies | [
"Tereza Pavlíková"
] | Same as the pdf. | [
"thinking for speaking",
"seeing for speaking",
"aspect",
"conceptualization",
"bilingualism"
] | https://openreview.net/pdf?id=Q3n9da9OQ2 | 9sI7i5XEXX | official_review | 1,736,280,054,602 | Q3n9da9OQ2 | [
"everyone"
] | [
"~Anna_Staňková1"
] | title: Great!
review: I think this is a well-structured abstract. In the first paragraph you write that "Many studies have found effects relevant to Slobin’s hypothesis" and then you cite a diploma thesis with an overview of the previous research. In this case I think it would be better to pick around 3 most influential studies from the overview and cite them directly. I would suggest adding some information about the statistical method you used and provide p-values when you are writing about the results not being statistically significant. |
9MvlUNVLvS | Afázie u mluvčích českého znakového jazyka | [
"Katerina Holubova"
] | Pozadí
Afázie, tedy porucha jazykových schopností způsobená získaným poškozením mozku, je dlouhodobě zkoumána u mluvčích mluvených jazyků (např. Duffau et al., 2005; Liben & Jarema, 2006; Lehečková, 2016). Po zásadním zlomu, kdy Stokoe (1960) poukázal na to, že i znakové jazyky mají definiční rysy spojované s lidskými jazyky, a nejsou tedy pouhým souborem gest, vznikly v 80. letech první kazuistiky zaměřující se i na znakové jazyky (např. Poizner, Bellugi, Iragui, 1984). Ty poukázaly na to, že poškození levé hemisféry skutečně ovlivňuje produkci a percepci amerického znakového jazyka, zatímco produkce gest obvykle zůstala neporušena. V návaznosti na tyto kazuistiky poté vzniklo několik dalších studií, které došly k podobným závěrům i v dalších národních znakových jazycích (např. Corina, 1998; Pickell et al., 2005).
I přesto, že od prvních zjištění uplynulo již čtyřicet let, dosud bylo publikováno pouze několik studií věnujících se jazykovým poruchám ve znakových jazycích. Všechny studie se navíc zabývaly zejména produkcí a percepcí takových jazykových prostředků, které se vyskytují i v mluvených jazycích. Znakové jazyky však disponují i prostředky, které v řadě mluvených jazyků nenalezneme (např. simultánní konstrukce nebo využívání prostoru pro gramatické účely) a u kterých lze předpokládat, že budou nějakým způsobem ovlivněny poškozením jazykových oblastí v mozku.
Kromě výše uvedených skutečností navíc dosud chybí jakýkoli popis projevů afázie u českého znakového jazyka (ČZJ), což má praktické důsledky i mimo lingvistické odvětví, včetně absence komplexního přístupu k diagnostice a terapii fatických poruch u uživatelů ČZJ.
Tato studie se zaměřuje na specifické projevy afázie u mluvčích českého znakového jazyka a kladla si za cíl popsat jednotlivé reprezentace afázie napříč různými jazykovými rovinami, porovnat je s dosavadními poznatky o afázii v cizích znakových jazycích a prozkoumat konkrétní jazykové prostředky, které ve znakových jazycích dosud ani v zahraničí nebyly prozkoumány.
Metodologie
Výzkumu se zúčastnili dva neslyšící probandi – rodilí mluvčí českého znakového jazyka se získaným poškozením mozku projevujícím se poruchami v oblasti produkce jazyka. Elicitace byla tvořena čtyřmi částmi:
1. Popis obrázkového příběhu „Frog, Where Are You?“.
2. Polostrukturovaný rozhovor zaměřený na elicitaci konkrétních jazykových prostředků, jako jsou specifické znaky, inkorporace a záporné či tázací věty.
3. Porovnání aktuálního jazykového projevu pacientů s jejich videozáznamy před onemocněním, což umožnilo analyzovat změny v produkci znaků.
4. Komunikace pacientů s jejich rodinnými příslušníky na téma Vánoce, ve které byla elicitace zaměřena na produkci minulého a budoucího času; sekundárně se sledovala i komunikace pacienta jako taková – interakce, upoutání pozornosti, náhradní komunikační strategie atd.
Výsledky a závěr
Analýza nasbíraného materiálu poukázala na to, že projevy afázie v ČZJ odpovídají zjištěním u jiných znakových jazyků. Na fonologické rovině byly například pozorovány záměny ve všech parametrech znaku, jako je tvar ruky, místo artikulace a pohyb (viz Obr. 1). Na morfologicko-syntaktické rovině bylo zaznamenáno chybné užívání shodových a prostorových sloves, včetně používání nevhodných klasifikátorů. Pacienti například použili sloveso v citátové formě nebo klasifikátor pro dvounohé bytosti při popisu čtyřnohých zvířat.
Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány. Zaznamenáno bylo například chybné používání mimiky pro gramatické účely, čímž docházelo ke změnám významu celého sdělení, a to i přesto, že s nelingvistickým vyjádřením emocí pomocí výrazu obličeje neměli pacienti potíže. Afázie měla vliv i na produkci simultánních konstrukcí, kdy pacienti tyto konstrukce vyjadřovali sekvenčním způsobem (Obr. 2).
Celkově byly afatické projevy u českého znakového jazyka podobné těm, které byly popsány u mluvených jazyků. Odlišnosti vycházely zejména z obecných rozdílů mezi znakovými a mluvenými jazyky, jako je jiná modalita obou jazyků (vizuomotorická vs. audioorální), strukturní odlišnost jazyků a výskyt některých jazykových prostředků, které v mluvených jazycích nenalezneme. | [
"afázie",
"český znakový jazyk",
"neurolingvistika",
"jazykové poruchy",
"neslyšící"
] | https://openreview.net/pdf?id=9MvlUNVLvS | jYn08USDBy | official_review | 1,736,279,404,058 | 9MvlUNVLvS | [
"everyone"
] | [
"~Anna_Staňková1"
] | title: Dobrá práce!
review: Abstrakt je dobře napsaný a strukturovaný. Abstrakt srozumitelně popisuje, co bylo cílem výzkumu, jaká byla metoda i jaké jsou výsledky. Na některých místech by se abstrakt dal trochu zkrátit, kdyby bylo potřeba dodržet nějaký limit slov/znaků. Myslím, že v úvodní části není potřeba vysvětlovat, co je afázie (to by měl recenzent - snad - vědět). Podobně bych si dokázala představit trochu úspornější popis předcházejícího výzkumu. Co se týče metodologie, je mi jasné, že pro takovou studii je velmi těžké sehnat potřebné participanty. Každopádně bych se pak v popisu výsledků vyvarovala silných závěrů jako "Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány.", jelikož se výsledky opírají o data jen od dvou participantů (ale je to jen věc formulace). |
9MvlUNVLvS | Afázie u mluvčích českého znakového jazyka | [
"Katerina Holubova"
] | Pozadí
Afázie, tedy porucha jazykových schopností způsobená získaným poškozením mozku, je dlouhodobě zkoumána u mluvčích mluvených jazyků (např. Duffau et al., 2005; Liben & Jarema, 2006; Lehečková, 2016). Po zásadním zlomu, kdy Stokoe (1960) poukázal na to, že i znakové jazyky mají definiční rysy spojované s lidskými jazyky, a nejsou tedy pouhým souborem gest, vznikly v 80. letech první kazuistiky zaměřující se i na znakové jazyky (např. Poizner, Bellugi, Iragui, 1984). Ty poukázaly na to, že poškození levé hemisféry skutečně ovlivňuje produkci a percepci amerického znakového jazyka, zatímco produkce gest obvykle zůstala neporušena. V návaznosti na tyto kazuistiky poté vzniklo několik dalších studií, které došly k podobným závěrům i v dalších národních znakových jazycích (např. Corina, 1998; Pickell et al., 2005).
I přesto, že od prvních zjištění uplynulo již čtyřicet let, dosud bylo publikováno pouze několik studií věnujících se jazykovým poruchám ve znakových jazycích. Všechny studie se navíc zabývaly zejména produkcí a percepcí takových jazykových prostředků, které se vyskytují i v mluvených jazycích. Znakové jazyky však disponují i prostředky, které v řadě mluvených jazyků nenalezneme (např. simultánní konstrukce nebo využívání prostoru pro gramatické účely) a u kterých lze předpokládat, že budou nějakým způsobem ovlivněny poškozením jazykových oblastí v mozku.
Kromě výše uvedených skutečností navíc dosud chybí jakýkoli popis projevů afázie u českého znakového jazyka (ČZJ), což má praktické důsledky i mimo lingvistické odvětví, včetně absence komplexního přístupu k diagnostice a terapii fatických poruch u uživatelů ČZJ.
Tato studie se zaměřuje na specifické projevy afázie u mluvčích českého znakového jazyka a kladla si za cíl popsat jednotlivé reprezentace afázie napříč různými jazykovými rovinami, porovnat je s dosavadními poznatky o afázii v cizích znakových jazycích a prozkoumat konkrétní jazykové prostředky, které ve znakových jazycích dosud ani v zahraničí nebyly prozkoumány.
Metodologie
Výzkumu se zúčastnili dva neslyšící probandi – rodilí mluvčí českého znakového jazyka se získaným poškozením mozku projevujícím se poruchami v oblasti produkce jazyka. Elicitace byla tvořena čtyřmi částmi:
1. Popis obrázkového příběhu „Frog, Where Are You?“.
2. Polostrukturovaný rozhovor zaměřený na elicitaci konkrétních jazykových prostředků, jako jsou specifické znaky, inkorporace a záporné či tázací věty.
3. Porovnání aktuálního jazykového projevu pacientů s jejich videozáznamy před onemocněním, což umožnilo analyzovat změny v produkci znaků.
4. Komunikace pacientů s jejich rodinnými příslušníky na téma Vánoce, ve které byla elicitace zaměřena na produkci minulého a budoucího času; sekundárně se sledovala i komunikace pacienta jako taková – interakce, upoutání pozornosti, náhradní komunikační strategie atd.
Výsledky a závěr
Analýza nasbíraného materiálu poukázala na to, že projevy afázie v ČZJ odpovídají zjištěním u jiných znakových jazyků. Na fonologické rovině byly například pozorovány záměny ve všech parametrech znaku, jako je tvar ruky, místo artikulace a pohyb (viz Obr. 1). Na morfologicko-syntaktické rovině bylo zaznamenáno chybné užívání shodových a prostorových sloves, včetně používání nevhodných klasifikátorů. Pacienti například použili sloveso v citátové formě nebo klasifikátor pro dvounohé bytosti při popisu čtyřnohých zvířat.
Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány. Zaznamenáno bylo například chybné používání mimiky pro gramatické účely, čímž docházelo ke změnám významu celého sdělení, a to i přesto, že s nelingvistickým vyjádřením emocí pomocí výrazu obličeje neměli pacienti potíže. Afázie měla vliv i na produkci simultánních konstrukcí, kdy pacienti tyto konstrukce vyjadřovali sekvenčním způsobem (Obr. 2).
Celkově byly afatické projevy u českého znakového jazyka podobné těm, které byly popsány u mluvených jazyků. Odlišnosti vycházely zejména z obecných rozdílů mezi znakovými a mluvenými jazyky, jako je jiná modalita obou jazyků (vizuomotorická vs. audioorální), strukturní odlišnost jazyků a výskyt některých jazykových prostředků, které v mluvených jazycích nenalezneme. | [
"afázie",
"český znakový jazyk",
"neurolingvistika",
"jazykové poruchy",
"neslyšící"
] | https://openreview.net/pdf?id=9MvlUNVLvS | iz5UVewf0P | official_review | 1,736,347,363,709 | 9MvlUNVLvS | [
"everyone"
] | [
"~Barbora_Genserová1"
] | title: Hezký abstrakt!
review: Text je srozumitelně napsaný, logicky a přehledně strukturovaný, metodologie podrobně vysvětlená.
Na abstrakt je text příliš dlouhý. Přispívají k tomu obrázky, které bychom ale mohli brát podobně jako glosy a nepočítat do rozsahu abstraktu. Naopak dobře zkrátit by šel úvod (lze se spolehnout na to, že jsou obecné informace v lingvistické komunitě známé, a pokud nejsou, čtenáři si je dohledají podle citovaných referencí).
Pro bezpečnější závěry by bylo dobré provést studii s více participanty, i když je mi jasné, že sehnat takovouto skupinu bude náročné. Je ale skvělé, že jsi měla k dispozici srovnání jazykového projevu před afázií se současnou produkcí. |
9MvlUNVLvS | Afázie u mluvčích českého znakového jazyka | [
"Katerina Holubova"
] | Pozadí
Afázie, tedy porucha jazykových schopností způsobená získaným poškozením mozku, je dlouhodobě zkoumána u mluvčích mluvených jazyků (např. Duffau et al., 2005; Liben & Jarema, 2006; Lehečková, 2016). Po zásadním zlomu, kdy Stokoe (1960) poukázal na to, že i znakové jazyky mají definiční rysy spojované s lidskými jazyky, a nejsou tedy pouhým souborem gest, vznikly v 80. letech první kazuistiky zaměřující se i na znakové jazyky (např. Poizner, Bellugi, Iragui, 1984). Ty poukázaly na to, že poškození levé hemisféry skutečně ovlivňuje produkci a percepci amerického znakového jazyka, zatímco produkce gest obvykle zůstala neporušena. V návaznosti na tyto kazuistiky poté vzniklo několik dalších studií, které došly k podobným závěrům i v dalších národních znakových jazycích (např. Corina, 1998; Pickell et al., 2005).
I přesto, že od prvních zjištění uplynulo již čtyřicet let, dosud bylo publikováno pouze několik studií věnujících se jazykovým poruchám ve znakových jazycích. Všechny studie se navíc zabývaly zejména produkcí a percepcí takových jazykových prostředků, které se vyskytují i v mluvených jazycích. Znakové jazyky však disponují i prostředky, které v řadě mluvených jazyků nenalezneme (např. simultánní konstrukce nebo využívání prostoru pro gramatické účely) a u kterých lze předpokládat, že budou nějakým způsobem ovlivněny poškozením jazykových oblastí v mozku.
Kromě výše uvedených skutečností navíc dosud chybí jakýkoli popis projevů afázie u českého znakového jazyka (ČZJ), což má praktické důsledky i mimo lingvistické odvětví, včetně absence komplexního přístupu k diagnostice a terapii fatických poruch u uživatelů ČZJ.
Tato studie se zaměřuje na specifické projevy afázie u mluvčích českého znakového jazyka a kladla si za cíl popsat jednotlivé reprezentace afázie napříč různými jazykovými rovinami, porovnat je s dosavadními poznatky o afázii v cizích znakových jazycích a prozkoumat konkrétní jazykové prostředky, které ve znakových jazycích dosud ani v zahraničí nebyly prozkoumány.
Metodologie
Výzkumu se zúčastnili dva neslyšící probandi – rodilí mluvčí českého znakového jazyka se získaným poškozením mozku projevujícím se poruchami v oblasti produkce jazyka. Elicitace byla tvořena čtyřmi částmi:
1. Popis obrázkového příběhu „Frog, Where Are You?“.
2. Polostrukturovaný rozhovor zaměřený na elicitaci konkrétních jazykových prostředků, jako jsou specifické znaky, inkorporace a záporné či tázací věty.
3. Porovnání aktuálního jazykového projevu pacientů s jejich videozáznamy před onemocněním, což umožnilo analyzovat změny v produkci znaků.
4. Komunikace pacientů s jejich rodinnými příslušníky na téma Vánoce, ve které byla elicitace zaměřena na produkci minulého a budoucího času; sekundárně se sledovala i komunikace pacienta jako taková – interakce, upoutání pozornosti, náhradní komunikační strategie atd.
Výsledky a závěr
Analýza nasbíraného materiálu poukázala na to, že projevy afázie v ČZJ odpovídají zjištěním u jiných znakových jazyků. Na fonologické rovině byly například pozorovány záměny ve všech parametrech znaku, jako je tvar ruky, místo artikulace a pohyb (viz Obr. 1). Na morfologicko-syntaktické rovině bylo zaznamenáno chybné užívání shodových a prostorových sloves, včetně používání nevhodných klasifikátorů. Pacienti například použili sloveso v citátové formě nebo klasifikátor pro dvounohé bytosti při popisu čtyřnohých zvířat.
Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány. Zaznamenáno bylo například chybné používání mimiky pro gramatické účely, čímž docházelo ke změnám významu celého sdělení, a to i přesto, že s nelingvistickým vyjádřením emocí pomocí výrazu obličeje neměli pacienti potíže. Afázie měla vliv i na produkci simultánních konstrukcí, kdy pacienti tyto konstrukce vyjadřovali sekvenčním způsobem (Obr. 2).
Celkově byly afatické projevy u českého znakového jazyka podobné těm, které byly popsány u mluvených jazyků. Odlišnosti vycházely zejména z obecných rozdílů mezi znakovými a mluvenými jazyky, jako je jiná modalita obou jazyků (vizuomotorická vs. audioorální), strukturní odlišnost jazyků a výskyt některých jazykových prostředků, které v mluvených jazycích nenalezneme. | [
"afázie",
"český znakový jazyk",
"neurolingvistika",
"jazykové poruchy",
"neslyšící"
] | https://openreview.net/pdf?id=9MvlUNVLvS | 6HNVAY75Do | official_review | 1,736,345,294,818 | 9MvlUNVLvS | [
"everyone"
] | [
"~Maria_Onoeva1"
] | title: Dobrá práce!
review: Tento abstrakt představuje malou studii zaměřenou na uživatele českého znakového jazyka s afázií. Studie se účastnili pouze dva respondenti, přesto výsledky korespondují s dosavadními poznatky o mluvených jazycích a zároveň přinášejí nové informace specifické pro znakové jazyky.
Mezi silné stránky abstraktu patří jasně formulovaná výzkumná otázka, logická struktura textu a přehledné zpracování vizuálních prvků. Pro zlepšení by bylo vhodné více se zaměřit na význam dosažených výsledků a jejich možné dopady na další výzkum.
Otázky:
- Plánuješ zapojit víc účastníků? Pokud ano, změnila bys něco ve své metodě? |
9MvlUNVLvS | Afázie u mluvčích českého znakového jazyka | [
"Katerina Holubova"
] | Pozadí
Afázie, tedy porucha jazykových schopností způsobená získaným poškozením mozku, je dlouhodobě zkoumána u mluvčích mluvených jazyků (např. Duffau et al., 2005; Liben & Jarema, 2006; Lehečková, 2016). Po zásadním zlomu, kdy Stokoe (1960) poukázal na to, že i znakové jazyky mají definiční rysy spojované s lidskými jazyky, a nejsou tedy pouhým souborem gest, vznikly v 80. letech první kazuistiky zaměřující se i na znakové jazyky (např. Poizner, Bellugi, Iragui, 1984). Ty poukázaly na to, že poškození levé hemisféry skutečně ovlivňuje produkci a percepci amerického znakového jazyka, zatímco produkce gest obvykle zůstala neporušena. V návaznosti na tyto kazuistiky poté vzniklo několik dalších studií, které došly k podobným závěrům i v dalších národních znakových jazycích (např. Corina, 1998; Pickell et al., 2005).
I přesto, že od prvních zjištění uplynulo již čtyřicet let, dosud bylo publikováno pouze několik studií věnujících se jazykovým poruchám ve znakových jazycích. Všechny studie se navíc zabývaly zejména produkcí a percepcí takových jazykových prostředků, které se vyskytují i v mluvených jazycích. Znakové jazyky však disponují i prostředky, které v řadě mluvených jazyků nenalezneme (např. simultánní konstrukce nebo využívání prostoru pro gramatické účely) a u kterých lze předpokládat, že budou nějakým způsobem ovlivněny poškozením jazykových oblastí v mozku.
Kromě výše uvedených skutečností navíc dosud chybí jakýkoli popis projevů afázie u českého znakového jazyka (ČZJ), což má praktické důsledky i mimo lingvistické odvětví, včetně absence komplexního přístupu k diagnostice a terapii fatických poruch u uživatelů ČZJ.
Tato studie se zaměřuje na specifické projevy afázie u mluvčích českého znakového jazyka a kladla si za cíl popsat jednotlivé reprezentace afázie napříč různými jazykovými rovinami, porovnat je s dosavadními poznatky o afázii v cizích znakových jazycích a prozkoumat konkrétní jazykové prostředky, které ve znakových jazycích dosud ani v zahraničí nebyly prozkoumány.
Metodologie
Výzkumu se zúčastnili dva neslyšící probandi – rodilí mluvčí českého znakového jazyka se získaným poškozením mozku projevujícím se poruchami v oblasti produkce jazyka. Elicitace byla tvořena čtyřmi částmi:
1. Popis obrázkového příběhu „Frog, Where Are You?“.
2. Polostrukturovaný rozhovor zaměřený na elicitaci konkrétních jazykových prostředků, jako jsou specifické znaky, inkorporace a záporné či tázací věty.
3. Porovnání aktuálního jazykového projevu pacientů s jejich videozáznamy před onemocněním, což umožnilo analyzovat změny v produkci znaků.
4. Komunikace pacientů s jejich rodinnými příslušníky na téma Vánoce, ve které byla elicitace zaměřena na produkci minulého a budoucího času; sekundárně se sledovala i komunikace pacienta jako taková – interakce, upoutání pozornosti, náhradní komunikační strategie atd.
Výsledky a závěr
Analýza nasbíraného materiálu poukázala na to, že projevy afázie v ČZJ odpovídají zjištěním u jiných znakových jazyků. Na fonologické rovině byly například pozorovány záměny ve všech parametrech znaku, jako je tvar ruky, místo artikulace a pohyb (viz Obr. 1). Na morfologicko-syntaktické rovině bylo zaznamenáno chybné užívání shodových a prostorových sloves, včetně používání nevhodných klasifikátorů. Pacienti například použili sloveso v citátové formě nebo klasifikátor pro dvounohé bytosti při popisu čtyřnohých zvířat.
Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány. Zaznamenáno bylo například chybné používání mimiky pro gramatické účely, čímž docházelo ke změnám významu celého sdělení, a to i přesto, že s nelingvistickým vyjádřením emocí pomocí výrazu obličeje neměli pacienti potíže. Afázie měla vliv i na produkci simultánních konstrukcí, kdy pacienti tyto konstrukce vyjadřovali sekvenčním způsobem (Obr. 2).
Celkově byly afatické projevy u českého znakového jazyka podobné těm, které byly popsány u mluvených jazyků. Odlišnosti vycházely zejména z obecných rozdílů mezi znakovými a mluvenými jazyky, jako je jiná modalita obou jazyků (vizuomotorická vs. audioorální), strukturní odlišnost jazyků a výskyt některých jazykových prostředků, které v mluvených jazycích nenalezneme. | [
"afázie",
"český znakový jazyk",
"neurolingvistika",
"jazykové poruchy",
"neslyšící"
] | https://openreview.net/pdf?id=9MvlUNVLvS | 5ph1J95hKb | official_review | 1,736,340,151,773 | 9MvlUNVLvS | [
"everyone"
] | [
"~Lucie_Jarůšková1"
] | title: Pěkný abstrakt!
review: Jedná se o velmi zajímavé téma, které je velmi potřebné zkoumat. Oceňuji vybrání klinické populace v českém znakovém jazyce, což je v nyní v daném oboru unikátní skupina a obecně srovnání projevů afázie u mluveného a znakového jazyka je chvályhodné.
Abstrakt je psán hezkou spisovnou češtinou, s jasnou strukturou. Nicméně si myslím, že je na konferenční poměry příliš rozsáhlý, za což může i použití odstavců či seznamů. Stálo by za to zkrátit především úvodní teoretickou část, která jde příliš do šířky a obecnosti. Působí spíše jako úvod článku (ačkoliv zajímavě prezentuje danou problematiku; především pro neznalce oboru poskytuje solidní základ).
Oceňuji přidání obrázků, především pro mluvčí mluveného jazyka slouží pro lepší představu. Závěry a výsledky bych interpretovala opatrněji z důvodu menšího počtu participantů.
Abstrakt mě velmi zaujal, výzkum by měl být rozhodně prezentovaný na nějaké konferenci a držím autorce palce s dalším bádáním. |
9MvlUNVLvS | Afázie u mluvčích českého znakového jazyka | [
"Katerina Holubova"
] | Pozadí
Afázie, tedy porucha jazykových schopností způsobená získaným poškozením mozku, je dlouhodobě zkoumána u mluvčích mluvených jazyků (např. Duffau et al., 2005; Liben & Jarema, 2006; Lehečková, 2016). Po zásadním zlomu, kdy Stokoe (1960) poukázal na to, že i znakové jazyky mají definiční rysy spojované s lidskými jazyky, a nejsou tedy pouhým souborem gest, vznikly v 80. letech první kazuistiky zaměřující se i na znakové jazyky (např. Poizner, Bellugi, Iragui, 1984). Ty poukázaly na to, že poškození levé hemisféry skutečně ovlivňuje produkci a percepci amerického znakového jazyka, zatímco produkce gest obvykle zůstala neporušena. V návaznosti na tyto kazuistiky poté vzniklo několik dalších studií, které došly k podobným závěrům i v dalších národních znakových jazycích (např. Corina, 1998; Pickell et al., 2005).
I přesto, že od prvních zjištění uplynulo již čtyřicet let, dosud bylo publikováno pouze několik studií věnujících se jazykovým poruchám ve znakových jazycích. Všechny studie se navíc zabývaly zejména produkcí a percepcí takových jazykových prostředků, které se vyskytují i v mluvených jazycích. Znakové jazyky však disponují i prostředky, které v řadě mluvených jazyků nenalezneme (např. simultánní konstrukce nebo využívání prostoru pro gramatické účely) a u kterých lze předpokládat, že budou nějakým způsobem ovlivněny poškozením jazykových oblastí v mozku.
Kromě výše uvedených skutečností navíc dosud chybí jakýkoli popis projevů afázie u českého znakového jazyka (ČZJ), což má praktické důsledky i mimo lingvistické odvětví, včetně absence komplexního přístupu k diagnostice a terapii fatických poruch u uživatelů ČZJ.
Tato studie se zaměřuje na specifické projevy afázie u mluvčích českého znakového jazyka a kladla si za cíl popsat jednotlivé reprezentace afázie napříč různými jazykovými rovinami, porovnat je s dosavadními poznatky o afázii v cizích znakových jazycích a prozkoumat konkrétní jazykové prostředky, které ve znakových jazycích dosud ani v zahraničí nebyly prozkoumány.
Metodologie
Výzkumu se zúčastnili dva neslyšící probandi – rodilí mluvčí českého znakového jazyka se získaným poškozením mozku projevujícím se poruchami v oblasti produkce jazyka. Elicitace byla tvořena čtyřmi částmi:
1. Popis obrázkového příběhu „Frog, Where Are You?“.
2. Polostrukturovaný rozhovor zaměřený na elicitaci konkrétních jazykových prostředků, jako jsou specifické znaky, inkorporace a záporné či tázací věty.
3. Porovnání aktuálního jazykového projevu pacientů s jejich videozáznamy před onemocněním, což umožnilo analyzovat změny v produkci znaků.
4. Komunikace pacientů s jejich rodinnými příslušníky na téma Vánoce, ve které byla elicitace zaměřena na produkci minulého a budoucího času; sekundárně se sledovala i komunikace pacienta jako taková – interakce, upoutání pozornosti, náhradní komunikační strategie atd.
Výsledky a závěr
Analýza nasbíraného materiálu poukázala na to, že projevy afázie v ČZJ odpovídají zjištěním u jiných znakových jazyků. Na fonologické rovině byly například pozorovány záměny ve všech parametrech znaku, jako je tvar ruky, místo artikulace a pohyb (viz Obr. 1). Na morfologicko-syntaktické rovině bylo zaznamenáno chybné užívání shodových a prostorových sloves, včetně používání nevhodných klasifikátorů. Pacienti například použili sloveso v citátové formě nebo klasifikátor pro dvounohé bytosti při popisu čtyřnohých zvířat.
Kromě zmíněných skutečností se ukázalo, že poškození jazykových oblastí v mozku ovlivňuje i další jazykové prostředky, které dosud nebyly prozkoumány. Zaznamenáno bylo například chybné používání mimiky pro gramatické účely, čímž docházelo ke změnám významu celého sdělení, a to i přesto, že s nelingvistickým vyjádřením emocí pomocí výrazu obličeje neměli pacienti potíže. Afázie měla vliv i na produkci simultánních konstrukcí, kdy pacienti tyto konstrukce vyjadřovali sekvenčním způsobem (Obr. 2).
Celkově byly afatické projevy u českého znakového jazyka podobné těm, které byly popsány u mluvených jazyků. Odlišnosti vycházely zejména z obecných rozdílů mezi znakovými a mluvenými jazyky, jako je jiná modalita obou jazyků (vizuomotorická vs. audioorální), strukturní odlišnost jazyků a výskyt některých jazykových prostředků, které v mluvených jazycích nenalezneme. | [
"afázie",
"český znakový jazyk",
"neurolingvistika",
"jazykové poruchy",
"neslyšící"
] | https://openreview.net/pdf?id=9MvlUNVLvS | 0l72JprvwF | official_review | 1,736,341,725,544 | 9MvlUNVLvS | [
"everyone"
] | [
"~Radek_Šimík1"
] | title: Velmi zajímavý abstrakt, trošku moc dlouhé pozadí
review: Pozadí je formulováno spíše jako úvod k článku než k abstraktu. Ideálně byste se měla "vejít" do délky, kterou má teď první odstavec. V závislosti na tom, na jakou konferenci byste abstrakt posílala, je například možné předpokládat znalost toho, co je to afázie, a není to tedy třeba vysvětlovat. Vypustit je též možné formulace typu "po zásadním zlomu...", navíc se opět jedná o něco, co je v široké lingvistické komunitě dobře známá věc, není to tedy třeba obhajovat nebo zmiňovat.
Zde jen stručný nástin toho, jak to může vypadat:
Poškožení levé periferie mozku může vést k afatickým projevům u mluvčích znakových jazyků (reference). Pro český znakový jazyk dosud výzkum o afatických projevech chybí, což má negativní důsledky ... Předložená studie si klade za cíl...
Obsahové doporučení: Asi by stálo za to získat i nějaká kontrolní data. Nejen od mluvčích před afázií (to je samozřejmě taky skvělé), ale na úplně stejných úlohách od mluvčích bez afázie.
Jinak výsledky a závěry jsou popsány velmi hezky. Bylo by skvělé, kdyby bylo možné výsledky aspoň bazálně kvantifikovat. Nyní není jasné, jestli se zmíněné obtíže projevily spíš izolovaně, nebo šlo o častý jev, a pak případně jak častý. (A i proto by bylo dobré mít data od kontrolní skupiny, aby bylo s čím srovnávat v těch specifických věcech.) |
zd1LhBG0EV | Distributed Constrained Multi-Agent Reinforcement Learning with Consensus and Networked Communication | [
"Santiago Amaya-Corredor",
"Miguel Calvo-Fullana",
"Anders Jonsson"
] | Our research addresses scalability and coordination challenges inherent to distributed multi-agent systems (MAS) executing under operation constraints. We introduce a novel Constrained Multi-Agent Reinforcement Learning (CMARL) algorithm that integrates a consensus mechanism to ensure agent coordination. Our decentralized approach allows each agent to independently optimize its local rewards while adhering to global constraints evaluated via secondary rewards. These secondary rewards act as a coupling mechanism, penalizing non-cooperative behaviors. Agents operate within a communication network modeled as an undirected graph, exchanging information solely with immediate neighbors to dynamically update dual variables. Our algorithm is validated through its application to the economic dispatch problem within smart grid management, demonstrating its scalability and practical utility in optimizing energy distribution under operational constraints. Experimental results show that our method effectively balances the global and local objectives, proving its robustness in real-world, distributed settings. Key contributions of this work include: (i) the development of a CMARL algorithm that achieves long-term constraint satisfaction and agent consensus, (ii) an enhanced scalability of policy training through problem factorization based on observed state distributions, and (iii) the successful application of our algorithm in a smart grid management use case, highlighting its practical applicability and effectiveness in managing distributed energy resources. | [
"Multi-Agent Reinforcement Learning",
"Constrained Reinforcement Learning",
"Consensus",
"Distributed Optimisation",
"Networked Communication"
] | https://openreview.net/pdf?id=zd1LhBG0EV | X1YIKOilRy | decision | 1,722,287,918,008 | zd1LhBG0EV | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
zP9hpDEzPq | Viability of Future Actions: Robust Reinforcement Learning via Entropy Regularization | [
"Pierre-François Massiani",
"Alexander von Rohr",
"Lukas Haverbeck",
"Sebastian Trimpe"
] | Despite the many recent advances in reinforcement learning (RL), the question of learning policies that robustly satisfy state constraints under disturbances remains open. This paper reveals how robustness arises naturally by combining two common practices in unconstrained RL: entropy regularization and constraints penalization. Our results provide a method to learn robust policies, model-free and with standard popular algorithms. We begin by showing how entropy regularization biases the constrained RL problem towards maximizing the number of future viable actions, which is a form of robustness. Then, we relax the safety constraints via penalties to obtain an unconstrained RL problem, which we show approximates its constrained counterpart arbitrarily closely. We support our findings with illustrative examples and on popular RL benchmarks. | [
"reinforcement learning",
"viability",
"safe reinforcement learning",
"robust learning"
] | https://openreview.net/pdf?id=zP9hpDEzPq | SIyH7OxeoW | decision | 1,722,287,919,539 | zP9hpDEzPq | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
zHt4K5zX4P | Dreaming of Many Worlds: Learning Contextual World Models aids Zero-Shot Generalization | [
"Sai Prasanna",
"Karim Farid",
"Raghu Rajan",
"André Biedenkapp"
] | Zero-shot generalization (ZSG) to unseen dynamics is a major challenge for creating generally capable embodied agents. To address the broader challenge, we start with the simpler setting of contextual reinforcement learning (cRL), assuming observability of the context values that parameterize the variation in the system's dynamics, such as the mass or dimensions of a robot, without making further simplifying assumptions about the observability of the Markovian state. Toward the goal of ZSG to unseen variation in context, we propose the contextual recurrent state-space model (cRSSM), which introduces changes to the world model of the Dreamer (v3) (Hafner et al., 2023). This allows the world model to incorporate context for inferring latent Markovian states from the observations and modeling the latent dynamics. Our experiments show that such systematic incorporation of the context improves the ZSG of the policies trained on the ``dreams'' of the world model. The code for all our experiments is available at https://anonymous.4open.science/r/dreaming_many_worlds. | [
"Zero-Shot Generalization",
"Model-based",
"Dreamer",
"Contextual Reinforcement Learning"
] | https://openreview.net/pdf?id=zHt4K5zX4P | 9p4DQp6SBM | decision | 1,722,287,918,140 | zHt4K5zX4P | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
zAgGM0nFKp | Recurrent Natural Policy Gradient for POMDPs | [
"Semih Cayci",
"Atilla Eryilmaz"
] | In this paper, we study a natural policy gradient method based on recurrent neural networks (RNNs) for partially-observable Markov decision processes, whereby RNNs are used for policy parameterization and policy evaluation to address curse of dimensionality in reinforcement learning for POMDPs. We present finite-time and finite-width analyses for both the critic (recurrent temporal difference learning), and correspondingly-operated recurrent natural policy gradient method in the near-initialization regime. Our analysis demonstrates the efficiency of RNNs for problems with short-term memory with explicit bounds on the required network widths and sample complexity, and points out the challenges in the case of long-term dependencies. | [
"natural policy gradient",
"partially-observable Markov decision processes",
"partial observability",
"policy optimization",
"actor-critic",
"temporal difference learning"
] | https://openreview.net/pdf?id=zAgGM0nFKp | sq4vHnF9Gk | decision | 1,722,287,919,806 | zAgGM0nFKp | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
yPnchV2nLq | Beyond Stationarity: Convergence Analysis of Stochastic Softmax Policy Gradient Methods | [
"Sara Klein",
"Simon Weissmann",
"Leif Döring"
] | Markov Decision Processes (MDPs) are a formal framework for modeling and solving sequential decision-making problems. In finite time horizons such problems are relevant for instance for optimal stopping or specific supply chain problems, but also in the training of large language models. In contrast to infinite horizon MDPs optimal policies are not stationary, policies must be learned for every single epoch. In practice all parameters are often trained simultaneously, ignoring the inherent structure suggested by dynamic programming. This paper introduces a combination of dynamic programming and policy gradient called dynamical policy gradient, where the parameters are trained backwards in time.
For the tabular softmax parametrisation we carry out the convergence analysis for simultaneous and dynamic policy gradient towards global optima, both in the exact and sampled gradient settings without regularisation. It turns out that the use of dynamic policy gradient training much better exploits the structure of finite-time problems which is reflected in improved convergence bounds. | [
"reinforcement learning",
"policy gradient",
"stochastic approximation",
"finite-time MDP"
] | https://openreview.net/pdf?id=yPnchV2nLq | TJvByLbURa | decision | 1,722,287,918,869 | yPnchV2nLq | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
yDicN3WVZ2 | Interpretable and Editable Programmatic Tree Policies for Reinforcement Learning | [
"Hector Kohler",
"Quentin Delfosse",
"Riad Akrour",
"Kristian Kersting",
"Philippe Preux"
] | Deep reinforcement learning agents are prone to goal misalignments. The black-box nature of their policies hinders the detection and correction of such misalignments, and the trust necessary for real-world deployment. So far, solutions learning interpretable policies are inefficient or require many human priors.
We propose INTERPRETER, a fast distillation method producing INTerpretable Editable tRee Programs for ReinforcEmenT lEaRning. We empirically demonstrate that INTERPRETER compact tree programs match oracles across a diverse set of sequential decision tasks and evaluate the impact of our design choices on interpretability and performances. We show that our policies can be interpreted and edited to correct misalignments on Atari games and to explain real farming strategies. | [
"Programmatic policies",
"Imitation",
"Interpretable RL",
"Explainability"
] | https://openreview.net/pdf?id=yDicN3WVZ2 | IprFsRWe5M | decision | 1,722,287,918,469 | yDicN3WVZ2 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
xqA3py089g | Reducing Blackwell and Average Optimality to Discounted MDPs via the Blackwell Discount Factor | [
"Julien Grand-Clément",
"Marek Petrik"
] | We introduce the Blackwell discount factor for Markov Decision Processes (MDPs). Classical objectives for MDPs include discounted, average, and Blackwell optimality. Many existing approaches to computing average-optimal policies solve for discount-optimal policies with a discount factor close to $1$, but they only work under strong or hard-to-verify assumptions such as unichain or ergodicity. We highlight the shortcomings of the classical definition of Blackwell optimality, which does not lead to simple algorithms for computing Blackwell-optimal policies and overlooks the pathological behaviors of optimal value functions with respect to the discount factors. To resolve this issue, we show that when the discount factor is larger than the Blackwell discount factor $\gamma_{\sf bw}$, all discount-optimal policies become Blackwell- and average-optimal, and we derive a general upper bound on $\gamma_{\sf bw}$. Our upper bound on $\gamma_{\sf bw}$, parametrized by the bit-size of the rewards and transition probabilities of the MDP instance, provides the first reduction from average and Blackwell optimality to discounted optimality, *without any assumptions*, along with new polynomial-time algorithms. Our work brings new ideas from polynomials and algebraic numbers to the analysis of MDPs. Our results also apply to robust MDPs, enabling the first algorithms to compute robust Blackwell-optimal policies. | [
"Markov decision processes",
"average optimality",
"Blackwell optimality"
] | https://openreview.net/pdf?id=xqA3py089g | 5E2JCwxRpQ | decision | 1,722,287,916,265 | xqA3py089g | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
x60cH4KDRv | Towards Enhancing Representations in Reinforcement Learning using Relational Structure | [
"Aditya Mohan",
"Marius Lindauer"
] | While Deep Reinforcement Learning has demonstrated promising results, its practical application remains limited due to brittleness in complex environments characterized by attributes such as high-dimensional observations, sparse rewards, partial observability, and changing dynamics. To overcome these challenges, we propose enhancing representation learning in RL by incorporating structural inductive biases through Graph Neural Networks (GNNs). Our approach leverages a structured GNN latent model to capture relational structures, thereby improving belief representation end-to-end. We validate our model’s benefits through empirical evaluation in selected challenging environments within the Minigrid suite, which offers relational complexity, against a baseline that uses a Multi-Layer Perceptron (MLP) as the latent model. Additionally, we explore the robustness of these representations in continually changing environments by increasing the size and adding decision points in the form of distractors. Through this analysis, we offer initial insights into the advantages of combining relational latent representations using GNNs for end-to-end representation learning in RL and pave the way for future methods of incorporating graph structure for representation learning in RL. | [
"Reinforcement Learning",
"Representation Learning"
] | https://openreview.net/pdf?id=x60cH4KDRv | jf9NFG7nn8 | decision | 1,722,287,920,136 | x60cH4KDRv | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
wAVoH6QN8z | Unbiased Policy Gradient with Random Horizon | [
"Rui Yuan",
"Andrii Tretynko",
"Simone Rossi",
"Thomas Hannagan"
] | Policy gradient (PG) methods are widely used in reinforcement learning. However, for infinite-horizon discounted reward settings, practical implementations of PG usually must rely on biased gradient estimators, due to the truncated finite-horizon sampling, which limits actual performance and hinders theoretical analysis. In this work, we introduce a new family of algorithms, __unbiased policy gradient__ (UPG), that enables unbiased gradient estimators by considering finite-horizon undiscounted rewards, where the horizon is randomly sampled from a geometric distribution $\mathrm{Geom}(1-\gamma)$ associated to the discount factor $\gamma$. Thanks to the absence of bias, UPG achieves the $\mathcal{O}(\epsilon^{-4})$ sample complexity to a stationary point, which is improved by $\mathcal{O}(\log\epsilon^{-1})$, compared to the one of the vanilla PG, and is met with fewer assumptions. Our work also provides a new angle on well-known algorithms such as Q-PGT and RPG. We recover the unbiased Q-PGT algorithm as a special case of UPG, allowing for its first sample complexity analysis. We further show that UPG can be extended to $\alpha$-UPG, a more generic class of PG algorithms which performs unbiased gradient estimators and notably admits RPG as a special case. The general sample complexity analysis of $\alpha$-UPG that we present enables to recover the convergence rates of RPG, also with tighter bounds. Finally, we propose and evaluate two new algorithms within the UPG family: unbiased GPOMDP (UGPOMDP) and $\alpha$-UGPOMDP. We show theoretically and empirically on four different environments that both UGPOMDP and $\alpha$-UGPOMDP outperform its known vanilla PG counterpart, GPOMDP. | [
"Theory for Reinforcement Learning",
"Discounted Markov Decision Process",
"Policy Optimization",
"Policy Gradient",
"Sample Complexity"
] | https://openreview.net/pdf?id=wAVoH6QN8z | 7ZxPaZyDIM | decision | 1,722,287,916,684 | wAVoH6QN8z | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
vYm3GOD8CG | Deterministic Exploration via Stationary Bellman Error Maximization | [
"Sebastian Griesbach",
"Carlo D'Eramo"
] | Exploration is a crucial and distinctive aspect of reinforcement learning (RL) that
remains a fundamental open problem. Several methods have been proposed to
tackle this challenge. Commonly used methods inject random noise directly into
the actions, indirectly via entropy maximization, or add intrinsic rewards that
encourage the agent to steer to novel regions of the state space. Another previously
seen idea is to use the Bellman error as a separate optimization objective for
exploration. In this paper, we introduce three modifications to stabilize the latter
and arrive at a deterministic exploration policy. Our separate exploration agent
is informed about the state of the exploitation, thus enabling it to account for
previous experiences. Further components are introduced to make the exploration
objective agnostic toward the episode length and to mitigate instability introduced
by far-off-policy learning. Our experimental results show that our approach can
outperform ε-greedy in dense and sparse reward settings. | [
"reinforcement learning; exploration; fingerprinting; maximum reward;"
] | https://openreview.net/pdf?id=vYm3GOD8CG | hnV5JAHe2d | decision | 1,722,287,918,323 | vYm3GOD8CG | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
vUuUPszknz | Model-Based Meta-Reinforcement Learning for Hyperparameter Optimization | [
"Jeroen Albrechts",
"Hugo Max MARTIN",
"Maryam Tavakol"
] | Hyperparameter Optimization (HPO) plays a significant role in enhancing the performance of machine learning models. However, as the size and complexity of (deep) neural architectures continue to increase, conducting HPO has become very expensive in terms of time and computational resources. Existing methods that automate this process still demand numerous evaluations to find the optimal hyperparameter configurations. In this paper, we present a novel approach based on model-based reinforcement learning to effectively improve sample efficiency while minimizing resource consumption. We formulate the HPO task as a Markov decision process and develop a predictive dynamics model for efficient policy optimization. Additionally, we employ the Deep Sets framework to encode the state space, which is then leveraged in meta-learning for transfer of knowledge across multiple datasets, enabling the model to quickly adapt to new datasets. Empirical studies demonstrate that our approach outperforms alternative techniques on publicly available datasets in terms of sample efficiency and accuracy. | [
"Hyperparameter optimization",
"model-based reinforcement learning",
"meta-learning",
"state representation"
] | https://openreview.net/pdf?id=vUuUPszknz | eBUsaWEzsk | decision | 1,722,287,917,779 | vUuUPszknz | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
vU0CMgRVEn | Bandit Pareto Set Identification in a Multi-Output Linear Model | [
"Cyrille Kone",
"Emilie Kaufmann",
"Laura Richert"
] | We study the Pareto Set Identification (PSI) problem in a structured multi-output linear bandit model. In this setting each arm is associated a feature vector belonging to $\mathbb{R}^h$ and its mean vector in $\mathbb{R}^d$ linearly depends on this feature vector through a common unknown matrix $\Theta \in \mathbb{R}^{h \times d}$. The goal is to identity the set of non-dominated arms by adaptively collecting samples from the arms. We introduce and analyze the first optimal design-based algorithms for PSI, providing nearly optimal guarantees in both the fixed-budget and the fixed-confidence settings. Notably, we show that the difficulty of these tasks mainly depends on the sub-optimality gaps of $h$ arms only.
Our theoretical results are supported by an extensive benchmark on synthetic and real-world datasets. | [
"bandit",
"pure exploration",
"bandit pareto set identification",
"best arm identification"
] | https://openreview.net/pdf?id=vU0CMgRVEn | 1K9cluV134 | decision | 1,722,287,917,493 | vU0CMgRVEn | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
uliCm8TnGO | Hadamard Representations: Augmenting Hyperbolic Tangents in RL | [
"Jacob Eeuwe Kooi",
"Mark Hoogendoorn",
"Vincent Francois-Lavet"
] | Activation functions are one of the key components of a deep neural network. The most commonly used activation functions can be classed into the category of continuously differentiable (e.g. tanh) and linear-unit functions (e.g. ReLU), both having their own strengths and drawbacks with respect to downstream performance and representation capacity through learning (e.g. measured by the number of dead neurons and the effective rank). In reinforcement learning, the performance of continuously differentiable activations often falls short as compared to linear-unit functions. We provide insights into the vanishing gradients associated with the former, and show that the dying neuron problem is not exclusive to ReLU's. To alleviate vanishing gradients and the resulting dying neuron problem occurring with continuously differentiable activations, we propose a Hadamard representation. Using deep Q-networks and proximal policy optimization in the Atari domain, we show faster learning, a reduction in dead neurons and increased effective rank. | [
"Representation Learning",
"Reinforcement Learning",
"Activations"
] | https://openreview.net/pdf?id=uliCm8TnGO | F7CVydrrgX | decision | 1,722,287,920,956 | uliCm8TnGO | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
uSheVlIgzc | Leveraging diverse offline data in POMDPs with unobserved confounders | [
"Oussama Azizi",
"Philip Boeken",
"Onno Zoeter",
"Frans A Oliehoek",
"Matthijs T. J. Spaan"
] | In many Reinforcement Learning (RL) applications, offline data is readily available before an algorithm is deployed. Often, however, data-collection policies have had access to information that is not recorded in the dataset, requiring the RL agent to take unobserved confounders into account. We focus on the setting where the confounders are i.i.d. and, without additional assumptions on the strength of the confounding, we derive tight bounds for the causal effects of the actions on the observations and reward. In particular, we show that these bounds are tight when we leverage multiple datasets collected from diverse behavioral policies. We incorporate these bounds into Posterior Sampling for Reinforcement Learning (PSRL) and demonstrate their efficacy experimentally. | [
"Reinforcement Learning",
"Causal Inference",
"POMDP",
"Latent Confounding"
] | https://openreview.net/pdf?id=uSheVlIgzc | waRJCKKydf | decision | 1,722,287,919,960 | uSheVlIgzc | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: We encourage the authors to clarify their contribution and positioning, as pointed out by reviewer A7hR. |
u3UN2sSpqK | Policy Gradient Methods with Adaptive Policy Spaces | [
"Gianmarco Tedeschi",
"Matteo Papini",
"Alberto Maria Metelli",
"Marcello Restelli"
] | Policy search is one of the most effective reinforcement learning classes of methods for solving continuous control tasks. These methodologies attempt to find a good policy for an agent by fixing a family of parametric policies and then searching directly for the parameters that optimize the long-term reward. However, this parametric policy space represents just a subset of all possible Markovian policies, and finding a good parametrization for a given task is a challenging problem in its own right, typically left to human expertise. In this paper, we propose a novel, model-free, adaptive-space policy search algorithm, GAPS (Gradient-based Adaptive Policy Search). We start from a simple policy space; then, based on the observations we receive from the unknown environment, we build a sequence of policy spaces of increasing complexity, which yield more sophisticated optimized policies at each epoch. The final result is a parametric policy whose structure (including the number of parameters) is fitted on the problem at hand without any prior knowledge of the task. Finally, our algorithm is tested on a selection of continuous control tasks, evaluating the sequence of policies so obtained, and comparing the results with traditional policy optimization methods that employ a fixed policy space. | [
"Reinforcement Learning",
"Policy Search",
"Policy Optimization",
"Adaptive",
"Policy Gradient",
"Policy Space"
] | https://openreview.net/pdf?id=u3UN2sSpqK | PARWK0E79s | decision | 1,722,287,916,467 | u3UN2sSpqK | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
twulWll5GA | Adversarial Contextual Bandits Go Kernelized | [
"Gergely Neu",
"Julia Olkhovskaya",
"Sattar Vakili"
] | We study a generalization of the problem of online learning in adversarial linear contextual bandits by incorporating loss functions that belong to a reproducing kernel Hilbert space, which allows for a more flexible modeling of complex decision-making scenarios. We propose a computationally efficient algorithm that makes use of a new optimistically biased estimator for the loss functions and achieves near-optimal regret guarantees under a variety of eigenvalue decay assumptions made on the underlying kernel. Specifically, under the assumption of polynomial eigendecay with exponent~$c>1$, the regret is $\tilde{O}(KT^{\frac{1}{2}\pa{1+\frac{1}{c}}})$, where $T$ denotes the number of rounds and $K$ the number of actions. Furthermore, when the eigendecay follows an exponential pattern, we achieve an even tighter regret bound of $\tilde{O}(\sqrt{T})$. These rates match the lower bounds in all special cases where lower bounds are known at all, and match the best known upper bounds available for the more well-studied stochastic counterpart of our problem. | [
"Contextual Bandits"
] | https://openreview.net/pdf?id=twulWll5GA | R4iouhePh2 | decision | 1,722,287,918,408 | twulWll5GA | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
tQMBxQZblv | Pessimistic Iterative Planning for Robust POMDPs | [
"Maris F. L. Galesloot",
"Marnix Suilen",
"Thiago D. Simão",
"Steven Carr",
"Matthijs T. J. Spaan",
"ufuk topcu",
"Nils Jansen"
] | Robust partially observable Markov decision processes (robust POMDPs) extend classical POMDPs to handle additional uncertainty on the transition and observation probabilities via so-called uncertainty sets. Policies for robust POMDPs must not only be memory-based to account for partial observability but also robust against model uncertainty to account for the worst-case instances from the uncertainty
sets. We propose the pessimistic iterative planning (PIP) framework, which finds robust memory-based policies for robust POMDPs. PIP alternates between two main steps: (1) selecting an adversarial (non-robust) POMDP via worst-case probability instances from the uncertainty sets; and (2) computing a finite-state controller (FSC) for this adversarial POMDP. We evaluate the performance of this FSC on the original robust POMDP and use this evaluation in step (1) to select the next adversarial POMDP. Within PIP, we propose the rFSCNet algorithm. In each iteration, rFSCNet finds an FSC through a recurrent neural network by using supervision policies optimized for the adversarial POMDP. The empirical evaluation in four benchmark environments showcases improved robustness against several baseline methods and competitive performance compared to a state-of-the-art robust POMDP solver. | [
"Robust POMDPs",
"Planning",
"Recurrent neural networks",
"Finite-state controllers",
"Model-based RL"
] | https://openreview.net/pdf?id=tQMBxQZblv | QU5X2jogcZ | decision | 1,722,287,920,241 | tQMBxQZblv | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
t25wwgD9D5 | Efficient Rate Optimal Regret for Adversarial Contextual MDPs Using Online Function Approximation | [
"Orin Levy",
"Alon Cohen",
"Asaf Cassel",
"Yishay Mansour"
] | We present the OMG-CMDP! algorithm for regret minimization in adversarial Contextual MDPs. The algorithm operates under the minimal assumptions of realizable function class and access to online least squares and log loss regression oracles. Our algorithm is efficient (assuming efficient online regression oracles), simple and robust to approximation errors.
It enjoys an $\widetilde{O}(H^2 \sqrt{ TH|S||A| ( \mathcal{R}_{TH}(\mathcal{O}) + H log(1/\delta)} )$ regret guarantee,
with $T$ being the number of episodes, $S$ the state space, $A$ the action space, $H$ the horizon.
In addition, $\mathcal{R}_{TH}( \mathcal{O} )$
is the sum of the square and log-loss regression oracles' regret, used to approximate the context-dependent rewards and dynamics, respectively. To the best of our knowledge, our algorithm is the first efficient rate optimal regret minimization algorithm for adversarial CMDPs that operates under the minimal standard assumption of online function approximation. | [
"Regret",
"Online function approximation",
"Contextual MDPs",
"adversarial RL"
] | https://openreview.net/pdf?id=t25wwgD9D5 | ZknU3Zd0Xh | decision | 1,722,287,916,423 | t25wwgD9D5 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
ruXNZsmZ0G | A Conservative Approach for Few-Shot Transfer in Off-Dynamics Reinforcement Learning | [
"Paul Daoudi",
"CHRISTOPHE PRIEUR",
"Bogdan Robu",
"Merwan Barlier",
"Ludovic Dos Santos"
] | Off-dynamics Reinforcement Learning (ODRL) seeks to transfer a policy from a source environment to a target environment characterized by distinct yet similar dynamics. In this context, traditional RL agents depend excessively on the dynamics of the source environment, resulting in the discovery of policies that excel in this environment but fail to provide reasonable performance in the target one. In the few-shot framework, a limited number of transitions from the target environment are introduced to facilitate a more effective transfer. Addressing this challenge, we propose a new approach inspired by recent advancements in Imitation Learning and conservative RL algorithms. The proposed method introduces a penalty to regulate the trajectories generated by the source-trained policy. We evaluate our method across various environments representing diverse off-dynamics conditions, where it demonstrates performance improvements compared to existing baselines across most tested scenarios. | [
"Reinforcement Learning",
"Simulator-based Reinforcement Learning",
"Off-Dynamics Reinforcement Learning"
] | https://openreview.net/pdf?id=ruXNZsmZ0G | 5IEIdEmqWS | decision | 1,722,287,920,483 | ruXNZsmZ0G | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
rXSgsdvpV9 | Value Improved Actor Critic Algorithms | [
"Yaniv Oren",
"Moritz Akiya Zanger",
"Pascal R. Van der Vaart",
"Matthijs T. J. Spaan",
"Wendelin Boehmer"
] | Many modern reinforcement learning algorithms build on the actor-critic (AC) framework: iterative improvement of a policy (the actor) using *policy improvement operators* and iterative approximation of the policy's value (the critic). In contrast, the popular value-based algorithm family employs improvement operators in the value update, to iteratively improve the value function directly. In this work, we propose a general extension to the AC framework that employs two separate improvement operators: one applied to the policy in the spirit of policy-based algorithms and one applied to the value in the spirit of value-based algorithms, which we dub Value-Improved AC (VI-AC). We design two practical VI-AC algorithms based in the popular online off-policy AC algorithms TD3 and DDPG. We evaluate VI-TD3 and VI-DDPG in the Mujoco benchmark and find that both improve upon or match the performance of their respective baselines in all environments tested. | [
"Actor Critic",
"Dynamic Programming",
"Policy Improvement",
"TD3",
"DDPG",
"Reinforcement Learning"
] | https://openreview.net/pdf?id=rXSgsdvpV9 | LNU3BPo9Wq | decision | 1,722,287,916,113 | rXSgsdvpV9 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
rHb7Tb5Bdl | Contextualized Hybrid Ensemble Q-learning: Learning Fast with Control Priors | [
"Emma Cramer",
"Bernd Frauenknecht",
"Ramil Sabirov",
"Sebastian Trimpe"
] | Combining Reinforcement Learning (RL) with a prior controller can yield the best out of two worlds: RL can solve complex nonlinear problems, while the control prior ensures safer exploration and speeds up training. Prior work largely blends both components with a fixed weight, neglecting that the RL agent's performance varies with the training progress and across regions in the state space. Therefore, we advocate for an adaptive strategy that dynamically adjusts the weighting based on the RL agent's current capabilities. We propose a new adaptive hybrid RL algorithm, Contextualized Hybrid Ensemble Q-learning (CHEQ). CHEQ combines three key ingredients: (i) a time-invariant formulation of the adaptive hybrid RL problem treating the adaptive weight as a context variable, (ii) a weight adaption mechanism based on the parametric uncertainty of a critic ensemble, and (iii) ensemble-based acceleration for data-efficient RL. Evaluating CHEQ on a car racing task reveals substantially stronger data efficiency, exploration safety, and transferability to unknown scenarios than state-of-the-art adaptive hybrid RL methods. | [
"hybrid reinforcement learning",
"residual reinforcement learning"
] | https://openreview.net/pdf?id=rHb7Tb5Bdl | PiEPBuoo3X | decision | 1,722,287,920,775 | rHb7Tb5Bdl | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
rGKoYKLe3O | Truly No-Regret Learning in Constrained MDPs | [
"Adrian Müller",
"Pragnya Alatur",
"Volkan Cevher",
"Giorgia Ramponi",
"Niao He"
] | Constrained Markov decision processes (CMDPs) are a common way to model safety constraints in reinforcement learning. State-of-the-art methods for efficiently solving CMDPs are based on primal-dual algorithms. For these algorithms, all currently known regret bounds allow for *error cancellations* --- one can compensate for a constraint violation in one round with a strict constraint satisfaction in another. This makes the online learning process unsafe since it only guarantees safety for the final (mixture) policy but not during learning. As Efroni et al. (2020) pointed out, it is an open question whether primal-dual algorithms can provably achieve sublinear regret if we do not allow error cancellations. In this paper, we give the first affirmative answer. We first generalize a result on last-iterate convergence of regularized primal-dual schemes to CMDPs with multiple constraints. Building upon this insight, we propose a model-based primal-dual algorithm to learn in an unknown CMDP. We prove that our algorithm achieves sublinear regret without error cancellations. | [
"Constrained MDPs",
"Online Learning",
"Regret Bounds"
] | https://openreview.net/pdf?id=rGKoYKLe3O | dvKvsxEOBQ | decision | 1,722,287,918,843 | rGKoYKLe3O | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
rDMHb3KHH4 | Direct Advantage Estimation in Partially Observable Environments | [
"Hsiao-Ru Pan",
"Bernhard Schölkopf"
] | Direct Advantage Estimation (DAE) was recently shown to improve sample-efficiency of deep reinforcement learning (deep RL) algorithms; however, DAE assumes full observability of the environment, which may be restrictive in realistic settings. In the present work, we first show that DAE can be extended to partially observable domains with minor modifications. Secondly, we address the increased computational cost due to the need to approximate the transition probabilities through the use of discrete latent space models. Finally, we empirically evaluate
the proposed method using the Arcade Learning Environments, and show that it is scalable and sample-efficient. | [
"POMDP",
"advantage function",
"deep RL",
"off-policy learning"
] | https://openreview.net/pdf?id=rDMHb3KHH4 | wek5uW6R0y | decision | 1,722,287,917,491 | rDMHb3KHH4 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
qteUVvGvFQ | Explore-Go: Leveraging Exploration for Generalisation in Deep Reinforcement Learning | [
"Max Weltevrede",
"Felix Kaubek",
"Matthijs T. J. Spaan",
"Wendelin Boehmer"
] | One of the remaining challenges in reinforcement learning is to develop agents that can generalise to novel scenarios they might encounter once deployed. This challenge is often framed in a multi-task setting where agents train on a fixed set of tasks and have to generalise to new tasks. Recent work has shown that in this setting increased exploration during training can be leveraged to increase the generalisation performance of the agent. This makes sense when the states encountered during testing can actually be explored during training. In this paper, we provide intuition why exploration can also benefit generalisation to states that cannot be explicitly encountered during training. Additionally, we propose a novel method Explore-Go that exploits this intuition by increasing the number of states on which the agent trains. Explore-Go effectively increases the starting state distribution of the agent and as a result can be used in conjunction with most existing on-policy or off-policy reinforcement learning algorithms. We show empirically that our method can increase generalisation performance in an illustrative environment and on the Procgen benchmark. | [
"deep reinforcement learning",
"exploration",
"generalization"
] | https://openreview.net/pdf?id=qteUVvGvFQ | 0rnRHLOyof | decision | 1,722,287,918,965 | qteUVvGvFQ | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
q75NXQdzdJ | Denoised Predictive Imagination: An Information-theoretic approach for learning World Models | [
"Vedant Dave",
"Elmar Rueckert"
] | Humans excel at isolating relevant information from noisy data to predict the behavior of dynamic systems, effectively disregarding non-informative, temporally-correlated noise. In contrast, existing reinforcement learning algorithms face challenges in generating noise-free predictions within high-dimensional, noise-saturated environments, especially when trained on world models featuring realistic background noise extracted from natural video streams. We propose a novel information-theoretic approach that learn world models based on minimising the past information and retaining maximal information about the future, aiming at simultaneously learning control policies and at producing denoised predictions. Utilizing Soft Actor-Critic agents augmented with an information-theoretic auxiliary loss, we validate our method's effectiveness on complex variants of the standard DeepMind Control Suite tasks, where natural videos filled with intricate and task-irrelevant information serve as a background. Experimental results demonstrate that our model outperforms nine state-of-the-art approaches in various settings where natural videos serve as dynamic background noise. Our analysis also reveals that all these methods encounter challenges in more complex environments. | [
"Information Theory",
"Distractors",
"Predictive Information",
"Information Bottleneck",
"Reinforcement Learning"
] | https://openreview.net/pdf?id=q75NXQdzdJ | DK0HyE9CuJ | decision | 1,722,287,919,523 | q75NXQdzdJ | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
oUbjwcEI15 | Scaling Value Iteration Networks to 5000 Layers for Extreme Long-Term Planning | [
"Yuhui Wang",
"Qingyuan Wu",
"Weida Li",
"Dylan R. Ashley",
"Francesco Faccio",
"Chao Huang",
"Jürgen Schmidhuber"
] | The Value Iteration Network (VIN) is an end-to-end differentiable architecture that performs value iteration on a latent MDP for planning in reinforcement learning (RL). However, VINs struggle to scale to long-term and large-scale planning tasks, such as navigating a $100\times 100$ maze---a task which typically requires thousands of planning steps to solve. We observe that this deficiency is due to two issues: the representation capacity of the latent MDP and the planning module's depth. We address these by augmenting the latent MDP with a dynamic transition kernel, dramatically improving its representational capacity, and, to mitigate the vanishing gradient problem, introducing an "adaptive highway loss" that constructs skip connections to improve gradient flow. We evaluate our method on both 2D maze navigation environments and the ViZDoom 3D navigation benchmark. We find that our new method, named Dynamic Transition VIN (DT-VIN), easily scales to 5000 layers and casually solves challenging versions of the above tasks. Altogether, we believe that DT-VIN represents a concrete step forward in performing long-term large-scale planning in RL environments. | [
"Value Iteration Networks",
"Long-term Planning",
"Reinforcement Learning",
"Deep Neural Network"
] | https://openreview.net/pdf?id=oUbjwcEI15 | Ttuh84T3Qe | decision | 1,722,287,918,994 | oUbjwcEI15 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
oDTJ40pAH0 | Offline RL via Feature-Occupancy Gradient Ascent | [
"Gergely Neu",
"Nneka Okolo"
] | We study offline Reinforcement Learning in large infinite-horizon discounted Markov Decision Processes (MDPs) when the reward and transition models are linearly realizable under a known feature map. Starting from the classic linear-program formulation of the optimal control problem in MDPs, we develop a new algorithm that performs a form of gradient ascent in the space of feature occupancies, defined as the expected feature vectors that can potentially be generated by executing policies in the environment. We show that the resulting simple algorithm satisfies strong computational and sample complexity guarantees, achieved under the least restrictive data coverage assumptions known in the literature. In particular, we show that the sample complexity of our method scales optimally with the desired accuracy level and depends on a weak notion of coverage that only requires the empirical feature covariance matrix to cover a single direction in the feature space (as opposed to covering a full subspace). Additionally, our method is easy to implement and requires no prior knowledge of the coverage ratio (or even an upper bound on it), which altogether make it the strongest known algorithm for this setting to date. | [
"Offline Reinforcement Learning",
"Linear MDPs",
"Provably efficient RL"
] | https://openreview.net/pdf?id=oDTJ40pAH0 | XbXTK1NZxv | decision | 1,722,287,915,733 | oDTJ40pAH0 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
nBbGOI8yVN | Feudal Graph Reinforcement Learning | [
"Tommaso Marzi",
"Arshjot Singh Khehra",
"Andrea Cini",
"Cesare Alippi"
] | Graph-based representations and message-passing modular policies constitute prominent approaches to tackling composable control problems in Reinforcement Learning (RL). However, as shown by recent graph deep learning literature, such local message-passing operators can create information bottlenecks and hinder global coordination. The issue becomes more serious in tasks requiring high-level planning. In this work, we propose a novel methodology, named Feudal Graph Reinforcement Learning (FGRL), that addresses such challenges by relying on hierarchical RL and a pyramidal message-passing architecture. In particular, FGRL defines a hierarchy of policies where high-level commands are propagated from the top of the hierarchy down through a layered graph structure. The bottom layers mimic the morphology of the physical system, while the upper layers correspond to higher-order sub-modules. The resulting agents are then characterized by a committee of policies where actions at a certain level set goals for the level below, thus implementing a hierarchical decision-making structure that can naturally implement task decomposition. We evaluate the proposed framework on a graph clustering problem and MuJoCo locomotion tasks; simulation results show that FGRL compares favorably against relevant baselines. Furthermore, an in-depth analysis of the command propagation mechanism provides evidence that the introduced message-passing scheme favors learning hierarchical decision-making policies. | [
"reinforcement learning",
"graph representation learning",
"hierarchical reinforcement learning"
] | https://openreview.net/pdf?id=nBbGOI8yVN | SSgdjYHqvw | decision | 1,722,287,916,301 | nBbGOI8yVN | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
medEh35RGy | Understanding the Gaps in Satisficing Bandits | [
"Chloé Rouyer",
"Ronald Ortner",
"Peter Auer"
] | In this work, we consider a variation of the stochastic multi-armed bandit problem in which the learner is not necessarily trying to compete with the best arm, whose performance is not known ahead of time, but is satisfied with playing any arm that performs above a known satisficing threshold $S$. Michel et al. (2023) considered as respective performance measure the \textit{satisficing regret}, that scales in terms of the gaps between the expected performance of an insufficient arm and the threshold $S$, rather than in terms of its gap with the best arm. While Michel et al. propose an algorithm that achieves time-independent satisficing regret, their results suffer when arms are too close to the threshold. Is this dependency unavoidable?
The first contribution of our work is to provide an alternative and more general lower bound for the $K$-armed satisficing bandit problem, which highlights how the position of the threshold compared to the arms affects the bound.
Then, we introduce an algorithm robust against unbalanced gaps, which enjoys a nearly matching time-independent upper bound. We also propose an alternative definition of the satisficing regret, which might be better tailored to measure algorithm performance in these difficult instances and derive a lower bound for this regret.
Finally, we include experiments to compare these different regret measures and our proposed algorithms empirically. | [
"Satisficing Bandits",
"Multiarmed Bandits",
"Online Learning"
] | https://openreview.net/pdf?id=medEh35RGy | nqtdD1KDpU | decision | 1,722,287,920,647 | medEh35RGy | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: Although the analysis seems incomplete (only the two-arms case is proven), the idea appears sound and well developed in the two-agents case. We encourage the authors to include a more complete discussion for the n-arms case in their final version and hope this will trigger interesting discussions during EWRL. |
mWrTA7LnSq | Rate-Optimal Policy Optimization for Linear Markov Decision Processes | [
"Uri Sherman",
"Alon Cohen",
"Tomer Koren",
"Yishay Mansour"
] | We study regret minimization in online episodic linear Markov Decision Processes, and propose a policy optimization algorithm that is computationally efficient, and obtains rate optimal $\widetilde O (\sqrt K)$ regret where $K$ denotes the number of episodes. Our work is the first to establish the optimal rate (in terms of~$K$) of convergence in the stochastic setting with bandit feedback using a policy optimization based approach, and the first to establish the optimal rate in the adversarial setup with full information feedback, for which no algorithm with an optimal rate guarantee was previously known. | [
"Reinforcement Learning",
"Function Approximation",
"Online Optimization"
] | https://openreview.net/pdf?id=mWrTA7LnSq | W0X4NCkbuY | decision | 1,722,287,917,183 | mWrTA7LnSq | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
mMYvXU2SXG | Sample-efficient reinforcement learning for environments with rare high-reward states | [
"Daniel G Mastropietro",
"Urtzi Ayesta",
"Matthieu Jonckheere"
] | We introduce FVAC (Fleming-Viot Actor-Critic), an algorithm for efficient learning of optimal policies in reinforcement learning problems with rare, high-reward states. FVAC uses Actor-Critic policy gradient, with the critic estimated via the so-called Fleming-Viot particle system, a stochastic process used to model population evolution which is able to boost the visit frequency of the rare states. This frequency boosting is obtained by forcing exploration outside a set of states identified as highly visited during an initial exploration of the environment. The only requirements of the method are that learning must be set under the average reward criterion, and that a black-box simulator or emulator can be run on the environment. We showcase the method’s performance in windy grid worlds, where a non-zero reward is only observed at a terminal cell, which is difficult to reach due to the wind. Our results show that FVAC learns significantly faster than standard reinforcement learning algorithms based on Monte-Carlo exploration with temporal difference learning. | [
"Fleming-Viot",
"Actor-Critic",
"policy gradient",
"stochastic optimisation"
] | https://openreview.net/pdf?id=mMYvXU2SXG | gkm7BASbO1 | decision | 1,722,287,920,693 | mMYvXU2SXG | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: We encourage the authors to clarify the reviewers' concern in their final version, in particular concerning the experimental validation. |
lrghaBT4MO | Linear Bandits with Memory | [
"Giulia Clerici",
"Pierre Laforgue",
"Nicolò Cesa-Bianchi"
] | Nonstationary phenomena, such as satiation effects in recommendations, have mostly been modeled using bandits with finitely many arms. However, the richer action space provided by linear bandits is often preferred in practice. In this work, we introduce a novel nonstationary linear bandit model, where current rewards are influenced by the learner's past actions in a fixed-size window. Our model, which recovers stationary linear bandits as a special case, leverages two parameters: the window size $m \ge 0$, and an exponent $\gamma$ that captures the rotting ($\gamma < 0)$ or rising ($\gamma > 0$) nature of the phenomenon. When both $m$ and $\gamma$ are known, we propose and analyze a variant of OFUL which minimizes regret against cyclic policies. By choosing the cycle length so as to trade-off approximation and estimation errors, we then prove a bound of order $\sqrt{d}\,(m+1)^{\frac{1}{2}+\max\{\gamma,0\}}\,T^{3/4}$ (ignoring log factors) on the regret against the optimal sequence of actions, where $T$ is the horizon and $d$ is the dimension of the linear action space. Through a bandit model selection approach, our results are then extended to the case where both $m$ and $\gamma$ are unknown. Finally, we complement our theoretical results with experiments comparing our approach to natural baselines. | [
"Non-stationary Multiarmed bandits",
"Online Learning"
] | https://openreview.net/pdf?id=lrghaBT4MO | 9vtjJBvSML | decision | 1,722,287,919,390 | lrghaBT4MO | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
lo6LdYbx7f | Robust Chain of Thoughts Preference Optimization | [
"Eugene Choi",
"Arash Ahmadian",
"Olivier Pietquin",
"Matthieu Geist",
"Mohammad Gheshlaghi Azar"
] | Learning from human preferences has become the dominant paradigm in RL fine-tuning of large language models (LLMs). In particular human preferences are often distilled in the form of a reward model. Then this reward model is used through online RL methods to fine-tune the LLM. Alternatively offline methods like direct preference optimization (DPO) and Identity Preference Optimization (IPO) use contrastive losses to optimize the LLM directly by increasing the gap between the log-likelihoods of preferred and dis-preferred completions. Despite their success, these methods all suffer from a fundamental problem that their optimal solution highly depends on (and heavily optimized for) the behavior policy that has generated the completions of the preferences dataset. Therefore the solution of the existing methods may be prone to out-of-distribution (OOD) tasks where the validation dataset is significantly different from the behavior policy. Here we address this challenge by proposing Robust Chain of Thoughts Optimization (RoCoTO) of preferences, a practical and mathematically principled offline framework for reinforcement learning from human feedback that is completely robust to the changes in the behavior model. The key idea of \rocoto is to cast the problem of learning from human preferences as a self-improving chain of thoughts (CoT) process in which the goal is to learn a policy that is nearly perfect in the sense that its generations can be only minimally improved through the best self-improving CoT model. We show that this idea can be mathematically expressed in terms of a min-max optimization objective that aims at joint optimization of chain-of-thoughts policy and the main generative policy in an adversarial fashion. The solution for this joint optimization problem is independent of the behavior policy and thus it is robust to the changes in the behavior model. We then show that this objective can be re-expressed in the form of a non-adversarial IPO (DPO)-style (offline) loss which can be optimized using standard supervised optimization techniques at scale without any need for reward model and online inference. We show the effectiveness of RoCoTO in solving TL;DR summarization task and show its superiority to the baseline IPO and DPO when evaluated on OOD XSUM. | [
"RLHF",
"preference learning",
"LLM alignment"
] | https://openreview.net/pdf?id=lo6LdYbx7f | rE7y3GtZgA | decision | 1,722,287,921,096 | lo6LdYbx7f | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
kVVMFlbaA5 | Dual-Force: Enhanced Offline Diversity Maximization under Imitation Constraints | [
"Pavel Kolev",
"Marin Vlastelica",
"Georg Martius"
] | While many algorithms for diversity maximization under imitation constraints are online in nature, many applications require offline algorithms without environment interactions. Tackling this problem in the offline setting, however, presents significant challenges that require non-trivial, multi-stage optimization processes with non-stationary rewards. In this work, we present a novel offline algorithm that enhances diversity using an objective based on Van der Waals (VdW) force and successor features, and eliminates the need to learn a previously used skill discriminator. Moreover, by conditioning the value function and policy on a pre-trained Functional Reward Encoding (FRE), our method allows for better handling of non-stationary rewards and provides zero-shot recall of all skills encountered during training, significantly expanding the set of skills learned in prior work. Consequently, our algorithm benefits from receiving a consistently strong diversity signal (VdW), and enjoys more stable and efficient training. We demonstrate the effectiveness of our method in generating diverse skills for two robotic tasks in simulation: locomotion of a quadruped and local navigation with obstacle traversal. | [
"skill discovery",
"diversity",
"imitation",
"Fenchel duality",
"constrained RL",
"Van der Waals force",
"successor features",
"functional reward encoding"
] | https://openreview.net/pdf?id=kVVMFlbaA5 | 4U0cJCq5tY | decision | 1,722,287,920,963 | kVVMFlbaA5 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: Reviewer 6Mr6 (and others) pointed out the limited empirical evaluation, which would qualify the paper for rejection at a venue with proceedings. However, the idea seems of value to the community, seems promising and is likely to trigger interesting discussions, both on the robotics topics and the imitation learning ones. Therefore the paper seems borderilne, despite the constrasted reviews. The program chairs decide on acceptance, based on the trust that the authors provide the promised empirical validation and include it in the paper's final version. |
jlnA0ZRfv1 | Regret Guarantees for Adversarial Contextual Bandits with Delayed Feedback | [
"Liad Erez",
"Orin Levy",
"Yishay Mansour"
] | In this paper we present regret minimization algorithms for the contextual multi-armed bandit (CMAB) problem in the presence of delayed feedback, a scenario where reward observations arrive with delays chosen by an adversary. We study two fundamental frameworks in terms of the function classes used to derive regret bounds for CMAB. Firstly, for a finite policy class $ \Pi $, we establish an optimal regret bound of $ O \left( \sqrt{KT \log |\Pi|} + \sqrt{D \log |\Pi|} \right) $, where $ K $ is the number of arms, $ T $ is the number of rounds, and $ D $ is the sum of delays. Secondly, assuming a finite contextual reward function class $ \mathcal{F} $ and access to an online least-square regression oracle $\mathcal{O}$ over $\mathcal{F}$, we achieve a regret bound of $\widetilde{O}(\sqrt{KT\cdot (\mathcal{R}_T(\mathcal{O})+\log (\delta^{-1}))} + \eta D + d_m)$ that holds with probability at least $1-\delta$, where $d_m$ is the maximal delay, $\mathcal{R}_T(\mathcal{O})$ is an upper bound on the oracle's regret and $\eta$ is a stability parameter associated with the oracle. | [
"Regret",
"Adversarial Delays",
"Online Learning",
"Contextual Bandits with Delays"
] | https://openreview.net/pdf?id=jlnA0ZRfv1 | AeX5wpc5Kb | decision | 1,722,287,921,128 | jlnA0ZRfv1 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: The reviewers seem to disagree about the correctness of the proofs but the overall work presented is valuable and the claims of the paper deserve discussion. Although it is possible this paper would not be accepted at more selective venues, we believe it is worth giving it a poster slot and strongly encourage the authors to clarify the proof of theorem 1 and formulation of theorem 6, so that the discussion is fruitful and the paper is improved. |
jaFhipqjxR | Thompson Sampling-like Algorithms for Stochastic Rising Rested Bandits | [
"Marco Fiandri",
"Alberto Maria Metelli",
"Francesco Trovò"
] | Stochastic rising rested bandit (SRRB) is a specific bandit setting where the arms' expected rewards increase as they are pulled. They model scenarios in which the performances of the different options grow as an effect of an underlying learning process (e.g., online model selection). Even if the bandit literature provides specifically crafted algorithms based on upper-confidence bound approaches for such a setting, no study about Thompson sampling-like algorithms has been performed. Indeed, the specific trend and the strong regularity of the expected rewards given by the SRRB setting suggest that specific instances may be tackled effectively using classical Thompson sampling or some adapted versions. This work provides a novel theoretical analysis of the regret that such algorithms suffer in SRRB. Our results show that, under specific assumptions on the reward functions, even the Thompson sampling-like algorithms achieve the no-regret property. | [
"Rising bandits",
"Regret minimization",
"Thompson sampling"
] | https://openreview.net/pdf?id=jaFhipqjxR | wSwEDAHdE4 | decision | 1,722,287,917,384 | jaFhipqjxR | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
hpREDov0N2 | EVaR Optimization in MDPs with Total Reward Criterion | [
"Xihong Su",
"Marek Petrik",
"Julien Grand-Clément"
] | The infinite-horizon discounted objective is popular in reinforcement learning, partly due to stationary optimal policies and convenient analysis based on contracting Bellman operators. Unfortunately, optimal policies must be history-dependent for most common coherent risk-averse discounted objectives, such as Value at Risk (VaR) and Conditional Value at Risk (CVaR). They also must be computed using complex state augmentation schemes. In this paper, we show that the total reward objective, under the Entropic Risk Measure (ERM) and Entropic Value at Risk (EVaR), can be optimized by a stationary policy, an essential property for practical implementations. In addition, an optimal policy can be efficiently computed using linear programming. Importantly, our results only require the relatively mild condition of transient MDPs and allow for both positive and negative rewards, unlike prior work requiring assumptions on the sign of the rewards. Our results suggest that the total reward criterion may be preferable to the discounted criterion in a broad range of risk-averse reinforcement learning problems. | [
"MDP",
"EVaR",
"Stationary policy",
"Total reward"
] | https://openreview.net/pdf?id=hpREDov0N2 | GdPFa0pOio | decision | 1,722,287,916,538 | hpREDov0N2 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
hWeto6nboZ | Directed Exploration in Reinforcement Learning from Linear Temporal Logic | [
"Marco Bagatella",
"Andreas Krause",
"Georg Martius"
] | Linear temporal logic (LTL) is a powerful language for task specification in reinforcement learning, as it allows describing objectives beyond the expressivity of conventional discounted return formulations. Nonetheless, recent works have shown that LTL formulas can be translated into a variable rewarding and discounting scheme, whose optimization produces a policy maximizing a lower bound on the probability of formula satisfaction. However, the synthesized reward signal remains fundamentally sparse, making exploration challenging. We aim to overcome this limitation, which can prevent current algorithms from scaling beyond low-dimensional, short-horizon problems. We show how better exploration can be achieved by further leveraging the LTL specification and casting its corresponding Limit Deterministic Büchi Automaton (LDBA) as a Markov reward process, thus enabling a form of high-level value estimation. By taking a Bayesian perspective over LDBA dynamics and proposing a suitable prior distribution, we show that the values estimated through this procedure can be treated as a shaping potential and mapped to informative intrinsic rewards. Empirically, we demonstrate applications of our method from tabular settings to high-dimensional continuous systems, which have so far represented a significant challenge for LTL-based reinforcement learning algorithms. | [
"reinforcement learning",
"linear temporal logic",
"exploration",
"deep reinforcement learning"
] | https://openreview.net/pdf?id=hWeto6nboZ | lJoeac3jTF | decision | 1,722,287,919,382 | hWeto6nboZ | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
gtjvYFtuzx | Louvain Skills: Building Multi-Level Skill Hierarchies in Reinforcement Learning | [
"Joshua Benjamin Evans",
"Özgür Şimşek"
] | What is a useful skill hierarchy for an autonomous agent? We propose an answer based on a graphical representation of how the interaction between an agent and its environment may unfold. Our approach uses modularity maximisation as a central organising principle to expose the structure of the interaction graph at multiple levels of abstraction. The result is a collection of skills that operate at varying time scales, organised into a hierarchy, where skills that operate over longer time scales are composed of skills that operate over shorter time scales. The entire skill hierarchy is generated automatically, with no human intervention, including the skills themselves (their behaviour, when they can be called, and when they terminate) as well as the hierarchical dependency structure between them. In a wide range of environments, this approach generates skill hierarchies that are intuitively appealing and that considerably improve the learning performance of the agent. | [
"Reinforcement Learning",
"Hierarchical Reinforcement Learning",
"RL",
"HRL",
"Skill Discovery",
"Skill Hierarchies",
"Graph-Based",
"Graphs",
"Graph Clustering",
"Graph Partitioning"
] | https://openreview.net/pdf?id=gtjvYFtuzx | z4Qk1WTBIA | decision | 1,722,287,917,267 | gtjvYFtuzx | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
fmZOGLHIo9 | Time-Constrained Robust MDPs | [
"Adil Zouitine",
"David Bertoin",
"Pierre Clavier",
"Matthieu Geist",
"Emmanuel Rachelson"
] | Robust reinforcement learning is essential for deploying reinforcement learning algorithms in real-world scenarios where environmental uncertainty predominates.
Traditional robust reinforcement learning often depends on rectangularity assumptions, where adverse probability measures of outcome states are assumed to be independent across different states and actions.
This assumption, rarely fulfilled in practice, leads to overly conservative policies.
To address this problem, we introduce a new time-constrained robust MDP (TC-RMDP) formulation that considers multifactorial, correlated, and time-dependent disturbances, thus more accurately reflecting real-world dynamics. This formulation goes beyond the conventional rectangularity paradigm, offering new perspectives and expanding the analytical framework for robust RL.
We propose three distinct algorithms, each using varying levels of environmental information, and evaluate them extensively on continuous control benchmarks.
Our results demonstrate that these algorithms yield an efficient tradeoff between performance and robustness, outperforming traditional deep robust RL methods in time-constrained environments while preserving robustness in classical benchmarks.
This study revisits the prevailing assumptions in robust RL and opens new avenues for developing more practical and realistic RL applications. | [
"Robust reinforcement learning"
] | https://openreview.net/pdf?id=fmZOGLHIo9 | ZGSZkWUmYl | decision | 1,722,287,920,453 | fmZOGLHIo9 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
fe1sebjJRA | Learning mirror maps in policy mirror descent | [
"Carlo Alfano",
"Sebastian Rene Towers",
"Silvia Sapora",
"Chris Lu",
"Patrick Rebeschini"
] | Policy Mirror Descent (PMD) is a popular framework in reinforcement learning, serving as a unifying perspective that encompasses numerous algorithms. These algorithms are derived through the selection of a mirror map and enjoy finite-time convergence guarantees. Despite its popularity, the exploration of PMD's full potential is limited, with the majority of research focusing on a particular mirror map---namely, the negative entropy---which gives rise to the renowned Natural Policy Gradient (NPG) method. It remains uncertain from existing theoretical studies whether the choice of mirror map significantly influences PMD's efficacy. In our work, we conduct empirical investigations to show that the conventional mirror map choice (NPG) often yields less-than-optimal outcomes across several standard benchmark environments. Using evolutionary strategies, we identify more efficient mirror maps that enhance the performance of PMD. We first focus on a tabular environment, i.e. Grid-World, where we relate existing theoretical bounds with the performance of PMD for a few standard mirror maps and the learned one. We then show that it is possible to learn a mirror map that outperforms the negative entropy in more complex environments, such as the MinAtar suite. Our results suggest that mirror maps generalize well across various environments, raising questions about how to best match a mirror map to an environment's structure and characteristics. | [
"Reinforcement learning",
"meta-learning",
"mirror descent"
] | https://openreview.net/pdf?id=fe1sebjJRA | Wu2IVq41Xs | decision | 1,722,287,918,739 | fe1sebjJRA | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
bHS9XCCquT | RRLS : Robust Reinforcement Learning Suite | [
"Adil Zouitine",
"David Bertoin",
"Pierre Clavier",
"Matthieu Geist",
"Emmanuel Rachelson"
] | Robust reinforcement learning is the problem of learning control policies that provide optimal worst-case performance against a span of adversarial environments.
It is a crucial ingredient for deploying algorithms in real-world scenarios with prevalent environmental uncertainties and has been a long-standing object of attention in the community, without a standardized set of benchmarks.
This contribution endeavors to fill this gap. We introduce the Robust Reinforcement Learning Suite (RRLS), a benchmark suite based on Mujoco environments.
RRLS provides six continuous control tasks with two types of uncertainty sets for training and evaluation.
Our benchmark aims to standardize robust reinforcement learning tasks, facilitating reproducible and comparable experiments, in particular those from recent state-of-the-art contributions, for which we demonstrate the use of RRLS.
It is also designed to be easily expandable to new environments.
The source code is available at \href{https://anonymous.url}{https://anonymous.url}. | [
"Robust reinforcement learning benchmark"
] | https://openreview.net/pdf?id=bHS9XCCquT | Gp52zYu8Nj | decision | 1,722,287,918,585 | bHS9XCCquT | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
bHCnkZgfcN | Epistemic Bellman Operators | [
"Pascal R. Van der Vaart",
"Matthijs T. J. Spaan",
"Neil Yorke-Smith"
] | Uncertainty quantification remains a difficult challenge in reinforcement learning. Several algorithms exist that successfully quantify uncertainty in a practical setting, however it is unclear whether these algorithms are theoretically sound and can be expected to converge. Furthermore, they seem to treat the uncertainty in the target parameters in different ways. In this work, we unify several practical algorithms into one theoretical framework by defining a new Bellman operator on distributions, and show that this Bellman operator is a contraction. Further, building on our theory, we modify PPO, a popular modern model-free algorithm, into an uncertainty-aware variant to showcase the general applicability of our main result. | [
"reinforcement learning",
"uncertainty",
"bellman operator",
"proximal policy optimization"
] | https://openreview.net/pdf?id=bHCnkZgfcN | KcIqJNGA2D | decision | 1,722,287,920,098 | bHCnkZgfcN | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: Reviewer N9si pointed out a missing span of the related literature. We encourage the authors to include it in their final version (and in future submissions). |
bD6RWdvC8I | Evidence on the regularization properties of Maximum-Entropy Reinforcement Learning | [
"Remy Hosseinkhan Boucher",
"Lionel Mathelin",
"Onofrio Semeraro"
] | The generalisation and robustness properties of policies learnt through Maximum-Entropy Reinforcement Learning are investigated on chaotic dynamical systems with Gaussian noise on the observable.
First, the robustness under noise contamination of the agent's observation of entropy regularised policies is observed.
Second, notions of statistical learning theory, such as complexity measures on the learnt model, are borrowed to explain and predict the phenomenon.
Results show the existence of a relationship between entropy-regularised policy optimisation and robustness to noise, which can be described by the chosen complexity measures. | [
"Maximum-Entropy Reinforcement Learning",
"Robustness",
"Complexity Measures",
"Flat Minima",
"Fisher Information",
"Regularisation"
] | https://openreview.net/pdf?id=bD6RWdvC8I | qRWARjxZ6a | decision | 1,722,287,921,382 | bD6RWdvC8I | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: As reviewer tWab points out, it is likely this paper would not be accepted as is at a more selective venue, with proceedings. Nonetheless, the reviewers seem to agree on the interest of the presented work, even though its maturity might be debatable, both in terms of vocabulary, state-of-the-art comparison or convincingness of the results. For this reason, we decide to allocate a poster to this paper and strongly encourage the authors to take the reviewers' feedback into account to improve their paper, and use it also to prepare a fruitful discussion around the poster to help this work reach a better level of maturity. |
ZQnGWThxyF | Bootstrapping Expectiles in Reinforcement Learning | [
"Pierre Clavier",
"Emmanuel Rachelson",
"Erwan Le Pennec",
"Matthieu Geist"
] | Many classic Reinforcement Learning (RL) algorithms rely on a Bellman operator, which involves an expectation over the next states, leading to the concept of bootstrapping. To introduce a form of pessimism, we propose to replace this expectation with an expectile. In practice, this can be very simply done by replacing the $L_2$ loss with a more general expectile loss for the critic. Introducing pessimism in RL is desirable for various reasons, such as tackling the overestimation problem (for which classic solutions are double Q-learning or the twin-critic approach of TD3) or robust RL (where transitions are adversarial). We study empirically these two cases. For the overestimation problem, we show that the proposed approach, \texttt{ExpectRL}, provides better results than a classic twin-critic. On robust RL benchmarks, involving changes of the environment, we show that our approach is more robust than classic RL algorithms. We also introduce a variation of \texttt{ExpectRL} combined with domain randomization which is competitive with state-of-the-art robust RL agents. Eventually, we also extend \texttt{ExpectRL} with a mechanism for choosing automatically the expectile value, that is the degree of pessimism. | [
"Expectile",
"Robust RL",
"Robust MDPs"
] | https://openreview.net/pdf?id=ZQnGWThxyF | T4khENdRUo | decision | 1,722,287,916,753 | ZQnGWThxyF | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
ZNiBfA4AjX | A Minimax-Bayes Approach to Ad Hoc Teamwork | [
"Victor Villin",
"Christos Dimitrakakis",
"Thomas Kleine Buening"
] | Learning policies for Ad Hoc Teamwork (AHT) is challenging. Most standard methods choose a specific distribution over training partners, which is assumed to mirror the distribution over partners after deployment. Moreover, they offer limited guarantees over worst-case performance. To tackle the issue, we propose using a worst-case prior distribution by adapting ideas from minimax-Bayes analysis to AHT.
We thereby explicitly account for our uncertainty about the partners at test time. Extensive experiments, including evaluations on coordination tasks from the Melting Pot suite, show our method's superior robustness compared to self-play, fictitious play, and best response learning w.r.t. policy populations. This highlights the importance of selecting an appropriate training distribution over teammates to achieve robustness in AHT. | [
"ad hoc teamwork",
"robust reinforcement learning",
"multi-agent reinforcement learning"
] | https://openreview.net/pdf?id=ZNiBfA4AjX | gvVMXfxjkw | decision | 1,722,287,919,173 | ZNiBfA4AjX | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
ZEnbCWsxoL | ARLBench: Flexible and Efficient Benchmarking for Hyperparameter Optimization in Reinforcement Learning | [
"Jannis Becktepe",
"Julian Dierkes",
"Carolin Benjamins",
"Aditya Mohan",
"David Salinas",
"Raghu Rajan",
"Frank Hutter",
"Holger Hoos",
"Marius Lindauer",
"Theresa Eimer"
] | Hyperparameters are a critical factor in reliably training well-performing reinforcement learning (RL) agents. Unfortunately, developing and evaluating automated approaches for tuning such hyperparameters is both costly and time-consuming. As a result, such approaches are often only evaluated on a single domain or algorithm, making comparisons difficult and limiting insights into their generalizability. We propose ARLBench, a benchmark for hyperparameter optimization (HPO) in RL that allows comparisons of diverse HPO approaches while being highly efficient in evaluation. To enable research into HPO in RL, even in settings with low compute resources, we select a representative subset of HPO tasks spanning a variety of algorithm and environment combinations. This selection allows for generating a performance profile of an automated RL (AutoRL) method using only a fraction of the compute previously necessary, enabling a broader range of researchers to work on HPO in RL. With the extensive and large-scale dataset on hyperparameter landscapes that our selection is based on, ARLBench is an efficient, flexible, and future-oriented foundation for research on AutoRL. Both the benchmark and the dataset are available at https://github.com/automl/arlbench. | [
"Reinforcement Learning",
"Benchmarks",
"Hyperparameter Selection",
"Decision and Control",
"AutoRL",
"AutoML"
] | https://openreview.net/pdf?id=ZEnbCWsxoL | VsjoFWF1Bj | decision | 1,722,287,917,760 | ZEnbCWsxoL | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
ZCiWNGU75p | Backward explanations via redefinition of predicates | [
"Léo Saulières",
"Martin C. Cooper",
"Florence Dupin de Saint-Cyr"
] | History eXplanation based on Predicates (HXP) studies the behavior of a Reinforcement Learning (RL) agent in a sequence of agent's interactions with the environment (a history), through the prism of an arbitrary predicate [20].
To this end, an action importance score is computed for each action in the history. The explanation consists in displaying the most important actions to the user.
As the calculation of an action's importance is #W[1]-hard, it is necessary for long histories to approximate the scores, at the expense of their quality.
We therefore propose a new HXP method, called Backward-HXP, to provide explanations for these histories without having to approximate scores.
Experiments show the ability of B-HXP to summarise long histories. | [
"Explainable Reinforcement Learning",
"Sequence Explanation",
"Importance Score"
] | https://openreview.net/pdf?id=ZCiWNGU75p | llok9ZmNHv | decision | 1,722,287,917,348 | ZCiWNGU75p | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Z6o8YjA96I | Adaptive Exploration for Data-Efficient General Value Function Evaluations | [
"Arushi Jain",
"Josiah P. Hanna",
"Doina Precup"
] | General Value Functions (GVFs) (Sutton et al., 2011) represent predictive knowledge in reinforcement learning. Each GVF computes the expected return for a given policy, based on a unique reward. Existing methods relying on fixed behavior policies or pre-collected data often face data efficiency issues when learning multiple GVFs in parallel using off-policy methods. To address this, we introduce $GVFExplorer$ which adaptively learns a single behavior policy that efficiently collects data for evaluating multiple GVFs in parallel. Our method optimizes the behavior policy by minimizing the total variance in return across GVFs, thereby reducing the required environmental interactions We use an existing temporal-difference-style variance estimator to approximate the return variance. We prove that each behavior policy update decreases the overall mean squared error in GVF predictions. We empirically show our method's performance in tabular and nonlinear function approximation settings, including Mujoco environments, with stationary and non-stationary reward signals, optimizing data usage and reducing prediction errors across multiple GVFs. | [
"general value functions",
"GVFs",
"multiple policy evaluations",
"exploration for GVFs",
"variance-minimization"
] | https://openreview.net/pdf?id=Z6o8YjA96I | EzwnW6wxSR | decision | 1,722,287,918,021 | Z6o8YjA96I | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Z5rhPej0V7 | CrossQ: Batch Normalization in Deep Reinforcement Learning for Greater Sample Efficiency and Simplicity | [
"Aditya Bhatt",
"Daniel Palenicek",
"Boris Belousov",
"Max Argus",
"Artemij Amiranashvili",
"Thomas Brox",
"Jan Peters"
] | Sample efficiency is a crucial problem in deep reinforcement learning. Recent algorithms, such as REDQ and DroQ, found a way to improve the sample efficiency by increasing the update-to-data (UTD) ratio to 20 gradient update steps on the critic per environment sample. However, this comes at the expense of a greatly increased computational cost. To reduce this computational burden, we introduce CrossQ: A lightweight algorithm for continuous control tasks that makes careful use of Batch Normalization and removes target networks to surpass the current state-of-the-art in sample efficiency while maintaining a low UTD ratio of 1. Notably, CrossQ does not rely on advanced bias-reduction schemes used in current methods. CrossQ's contributions are threefold: (1) it matches or surpasses current state-of-the-art methods in terms of sample efficiency, (2) it substantially reduces the computational cost compared to REDQ and DroQ, (3) it is easy to implement, requiring just a few lines of code on top of SAC. | [
"reinforcement learning",
"deep learning"
] | https://openreview.net/pdf?id=Z5rhPej0V7 | vdtvLcvbRl | decision | 1,722,287,918,534 | Z5rhPej0V7 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
YBKCJUa7Dw | An Attentive Approach for Building Partial Reasoning Agents from Pixels | [
"Safa Alver",
"Doina Precup"
] | We study the problem of building reasoning agents that are able to generalize in an effective manner. Towards this goal, we propose an end-to-end approach for building model-based reinforcement learning agents that dynamically focus their reasoning to the relevant aspects of the environment: after automatically identifying the distinct aspects of the environment, these agents dynamically filter out the relevant ones and then pass them to their simulator to perform partial reasoning. Unlike existing approaches, our approach works with pixel-based inputs and it allows for interpreting the focal points of the agent. Our quantitative analyses show that the proposed approach allows for effective generalization in high-dimensional domains with raw observational inputs. We also perform ablation analyses to validate our design choices. Finally, we demonstrate through qualitative analyses that our approach actually allows for building agents that focus their reasoning on the relevant aspects of the environment. | [
"Model-based reinforcement learning",
"partial reasoning"
] | https://openreview.net/pdf?id=YBKCJUa7Dw | aRvi3G9ItK | decision | 1,722,287,919,092 | YBKCJUa7Dw | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
X3i12AKYJn | Curricula for Learning Robust Policies with Factored State Representations in Changing Environments | [
"Panayiotis Panayiotou",
"Özgür Şimşek"
] | Robust policies enable reinforcement learning agents to effectively adapt to and operate in unpredictable, dynamic, and ever-changing real-world environments. Factored representations, which break down complex state and action spaces into distinct components, can improve generalization and sample efficiency in policy learning. In this paper, we explore how the curriculum of an agent using a factored state representation affects the robustness of the learned policy. We experimentally demonstrate three simple curricula, such as varying only the variable of highest regret between episodes, that can significantly enhance policy robustness, offering practical insights for reinforcement learning in complex environments. | [
"Reinforcement Learning",
"Factored Representations",
"Policy Robustness",
"Curriculum Learning"
] | https://openreview.net/pdf?id=X3i12AKYJn | mdHvMRUbEB | decision | 1,722,287,920,266 | X3i12AKYJn | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
WhtVbN1Xjb | Preference Elicitation for Offline Reinforcement Learning | [
"Alizée Pace",
"Bernhard Schölkopf",
"Gunnar Ratsch",
"Giorgia Ramponi"
] | Applying reinforcement learning (RL) to real-world problems is often made challenging by the inability to interact with the environment and the difficulty of designing reward functions. Offline RL addresses the first challenge by considering access to an offline dataset of environment interactions labeled by the reward function. In contrast, Preference-based RL does not assume access to the reward function and learns it from preferences, but typically requires an online interaction with the environment. We bridge the gap between these frameworks by exploring efficient methods for acquiring preference feedback in a fully offline setup. We propose Sim-OPRL, an offline preference-based reinforcement learning algorithm, which leverages a learned environment model to elicit preference feedback on simulated rollouts. Drawing on insights from both the offline RL and the preference-based RL literature, our algorithm employs a pessimistic approach for out-of-distribution data, and an optimistic approach for acquiring informative preferences about the optimal policy. We provide theoretical guarantees regarding the sample complexity of our approach, dependent on how well the offline data covers the optimal policy. Finally, we demonstrate the empirical performance of Sim-OPRL in different environments. | [
"preference-based reinforcement learning",
"offline reinforcement learning",
"active learning",
"model-based reinforcement learning"
] | https://openreview.net/pdf?id=WhtVbN1Xjb | Yz3bHzFB3r | decision | 1,722,287,915,958 | WhtVbN1Xjb | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Vz99GHBGRO | Isoperimetry is All We Need: Langevin Posterior Sampling for RL | [
"Emilio Jorge",
"Christos Dimitrakakis",
"Debabrota Basu"
] | In Reinforcement Learning theory, we often assume restrictive assumptions, like linearity and RKHS structure on the model, or Gaussianity and log-concavity of the posteriors over models, to design an algorithm with provably sublinear regret. But RL in practice is known to work for wider range of distributions and models. Thus, we study whether we can design efficient low-regret RL algorithms for any isoperimetric distribution, which includes and extends the standard setups in the literature to non-log-concave and perturbed distributions. Specifically, we show that the well-known PSRL (Posterior Sampling-based RL) algorithm yields sublinear regret if the sequence of posterior distributions satisfy the Log-Sobolev Inequality (LSI), which is a form of isoperimetry, with linearly growing constants. Further, for the cases where we cannot compute or sample from an exact posterior, we propose a Langevin sampling-based algorithm design scheme, namely LaPSRL. We show that LaPSRL also achieves order optimal regret if the posteriors satisfy LSI. Finally, we deploy a version of LaPSRL with a Langevin sampling algorithms, SARAH-LD. We numerically demonstrate their performances in different bandit and MDP environments. Experimental results validate the generality of LaPSRL across environments and its competetive performance with respect to the baselines. | [
"Reinforcement learning theory",
"regret analysis",
"Langevin",
"log Sobolev inequality",
"PSRL",
"MDP"
] | https://openreview.net/pdf?id=Vz99GHBGRO | ga3a8mt1tu | decision | 1,722,287,916,871 | Vz99GHBGRO | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Vz03RUmrd1 | Individual Regret in Cooperative Stochastic Multi-Armed Bandits over Communication Graph | [
"Idan Barnea",
"Tal Lancewicki",
"Yishay Mansour"
] | We study the regret in stochastic Multi-Armed Bandits (MAB) with multiple agents that communicate over an arbitrary connected communication graph.
We show a near-optimal individual regret bound of $\tilde{O}(\sqrt{AT/m}+A)$, where $A$ is the number of actions, $T$ the time horizon, and $m$ the number of agents.
In particular, assuming a sufficient number of agents, we achieve a regret bound of $\tilde{O}(A)$, which is independent of the sub-optimality gaps and depends only logarithmically on the time horizon.
To the best of our knowledge, our study is the first to show an individual regret bound in cooperative stochastic MAB that is independent of the graph's diameter and applicable to non-fully-connected communication graphs. | [
"Multi-Armed Bandit",
"Cooperative Stochastic MAB",
"Regret Minimization",
"Multi-player MAB",
"Multi-agent Multi-Armed Bandit"
] | https://openreview.net/pdf?id=Vz03RUmrd1 | O2HskkgUfy | decision | 1,722,287,919,825 | Vz03RUmrd1 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: We encourage the authors to better position their work to the paper quoted by reviewer ikFn (as it is a JMLR 2022 paper) and to enhance their work with the perspectives suggested by reviewer W1Jt. |
Vx2ETvHId8 | A Look at Value-Based Decision-Time vs. Background Planning Methods Across Different Settings | [
"Safa Alver",
"Doina Precup"
] | In model-based reinforcement learning (RL), an agent can leverage a learned model to improve its way of behaving in different ways. Two of the prevalent ways to do this are through decision-time and background planning methods. In this study, we are interested in understanding how the value-based versions of these two planning methods will compare against each other across different settings. Towards this goal, we first consider the simplest instantiations of value-based decision-time and background planning methods and provide theoretical results on which one will perform better in the regular RL and transfer learning settings. Then, we consider the modern instantiations of them and provide hypotheses on which one will perform better in the same settings. Finally, we perform illustrative experiments to validate these theoretical results and hypotheses. Overall, our findings suggest that even though value-based versions of the two planning methods perform on par in their simplest instantiations, the modern instantiations of value-based decision-time planning methods can perform on par or better than the modern instantiations of value-based background planning methods in both the regular RL and transfer learning settings. | [
"Model-based reinforcement learning"
] | https://openreview.net/pdf?id=Vx2ETvHId8 | tBa8TL7gLL | decision | 1,722,287,920,585 | Vx2ETvHId8 | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: Although this is a borderline paper for EWRL, we believe it can lead to fruitful discussion during the workshop. We very strongly encourage the authors to address the reviewers' concerns, clarify the overlap with previous work by Alver and Precup, release their code, and provide the Dyna implementation as suggested by pjXR. |
VqEZ7qEG7P | Learning to Steer Markovian Agents under Model Uncertainty | [
"Jiawei Huang",
"Vinzenz Thoma",
"Zebang Shen",
"Heinrich H. Nax",
"Niao He"
] | We study reward design for steering multi-agent systems towards desired equilibria \emph{without} prior knowledge of the agents' underlying policy learning dynamics model. We introduce a model-based non-episodic Reinforcement Learning (RL) formulation for our steering problem. To handle the model uncertainty, we contribute a new optimization objective targeting at learning a \emph{history-dependent} steering strategy, and establish guarantees on its optimal solution. Theoretically, we identify conditions for the existence of steering strategies to guide agents sufficiently close to the desired policies. Complementing our theoretical contributions, we provide empirically tractable algorithms to approximately solve our objective, which effectively tackles the challenge in efficiently learning history-dependent strategies. We demonstrate the efficacy of our algorithms through empirical evaluations. | [
"Markov Games",
"Steering Learning Dynamics",
"Mechanism Design",
"Non-Episodic Reinforcement Learning"
] | https://openreview.net/pdf?id=VqEZ7qEG7P | OSOUvktbaU | decision | 1,722,287,915,831 | VqEZ7qEG7P | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
VRJaXWiu7j | Trust the Model Where It Trusts Itself - Model-Based Actor-Critic with Uncertainty-Aware Rollout Adaption | [
"Bernd Frauenknecht",
"Artur Eisele",
"Devdutt Subhasish",
"Friedrich Solowjow",
"Sebastian Trimpe"
] | Dyna-style model-based reinforcement learning (MBRL) combines model-free agents with predictive transition models through model-based rollouts. This combination raises a critical question: “When to trust your model?”; i.e., which rollout length results in the model providing useful data? Janner et al. (2019) address this question by gradually increasing rollout lengths throughout the training. While theoretically tempting, uniform model accuracy is a fallacy that collapses at the latest when extrapolating. Instead, we propose asking the question “Where to trust your model?”. Using inherent model uncertainty to consider local accuracy, we obtain the Model-Based Actor-Critic with Uncertainty-Aware Rollout Adaption (MACURA) algorithm. We propose an easy-to-tune rollout mechanism and demonstrate substantial improvements in data efficiency and performance compared to state-of-the-art deep MBRL methods on the MuJoCo benchmark. | [
"Model-based Reinforcement Learning",
"Model Uncertainty Estimation",
"Rollout Length Scheduling"
] | https://openreview.net/pdf?id=VRJaXWiu7j | pADD8EFS4c | decision | 1,722,287,919,432 | VRJaXWiu7j | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
SkSC03bV1F | Stochastic Q-learning for Large Discrete Action Spaces | [
"Fares Fourati",
"Vaneet Aggarwal",
"Mohamed-Slim Alouini"
] | In complex environments with large discrete action spaces, effective decision-making is critical in reinforcement learning (RL). Despite the widespread use of value-based RL approaches like Q-learning, they come with a computational burden, necessitating the maximization of a value function over all actions in each iteration. This burden becomes particularly challenging when addressing large-scale problems and using deep neural networks as function approximators.
In this paper, we present stochastic value-based RL approaches which, in each iteration, as opposed to optimizing over the entire set of $n$ actions, only consider a variable stochastic set of a sublinear number of actions, possibly as small as $\mathcal{O}(\log(n))$. The presented stochastic value-based RL methods include, among others, Stochastic Q-learning, StochDQN, and StochDDQN, all of which integrate this stochastic approach for both value-function updates and action selection. The theoretical convergence of Stochastic Q-learning is established, while an analysis of stochastic maximization is provided. Moreover, through empirical validation, we illustrate that the various proposed approaches outperform the baseline methods across diverse environments, including different control problems, achieving near-optimal average returns in significantly reduced time. | [
"Q-learning",
"Q-networks",
"Value-based RL",
"Reinforcement Learning",
"Large Discrete Action Spaces"
] | https://openreview.net/pdf?id=SkSC03bV1F | lCYDNwmumy | decision | 1,722,287,919,678 | SkSC03bV1F | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
SYV6AlWh9P | Revisiting On-Policy Deep Reinforcement Learning | [
"Mahdi Kallel",
"Samuele Tosatto",
"Carlo D'Eramo"
] | On-policy Reinforcement Learning (RL) offers several desirable properties, including more stable learning, less frequent policy changes, and the capacity to evaluate a policy's return during the learning process. Despite the considerable success of recent off-policy methods, their on-policy counterparts continue to lag in terms of asymptotic performance and sample efficiency. Proximal Policy Optimization (PPO) remains the de facto standard, despite its complexity and demonstrated sensitivity to hyperparameters. In this work, we introduce On-Policy Soft Actor-Critic (ON-SAC), a methodical adaptation of the Soft Actor-Critic (SAC) algorithm tailored for the on-policy setting. Our approach begins with the observation that the current on-policy algorithms do not use true on-policy gradients. We build on this observation to offer founded remedies for this problem. Our algorithm establishes a new state-of-the-art for deep on-policy RL, while simplifying the process by eliminating the need for trust-region methods and intricate critic learning schemes. | [
"Reinforcement learning",
"on-policy",
"maximum entropy RL"
] | https://openreview.net/pdf?id=SYV6AlWh9P | 9hd8CCu0Lv | decision | 1,722,287,919,311 | SYV6AlWh9P | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
SJLci0mRYa | Interpreting Reinforcement Learning with Shapley Values | [
"Daniel Beechey",
"Thomas M. S. Smith",
"Özgür Şimşek"
] | For reinforcement learning systems to be widely adopted, their users must understand and trust them. We present a theoretical analysis of explaining reinforcement learning using Shapley values, following a principled approach from game theory for identifying the contribution of individual players to the outcome of a cooperative game. We call this general framework Shapley Values for Explaining Reinforcement Learning (SVERL). Our analysis exposes the limitations of earlier uses of Shapley values in reinforcement learning. We then develop an approach that uses Shapley values to explain agent performance. In a variety of domains, SVERL produces meaningful explanations that match and supplement human intuition. | [
"ICML",
"machine learning",
"reinforcement learning",
"explainable AI",
"Shapley values",
"SVERL"
] | https://openreview.net/pdf?id=SJLci0mRYa | eOmJvtIVlI | decision | 1,722,287,917,226 | SJLci0mRYa | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
RlfCQ4cvmd | Earth Observation Satellite Scheduling with Graph Neural Networks | [
"Guillaume Infantes",
"Antoine Jacquet",
"Emmanuel Benazera",
"Stéphanie Roussel",
"Nicolas Meuleau",
"Vincent Baudoui",
"Jonathan Guerra"
] | The Earth Observation Satellite Planning (EOSP) is a difficult optimization prob-
lem with considerable practical interest. A set of requested observations must
be scheduled on an agile Earth observation satellite while respecting constraints
on their visibility window, as well as maneuver constraints that impose varying
delays between successive observations. In addition, the problem is largely over-
subscribed: there are much more candidate observations than what can possibly
be achieved. Therefore, one must select the set of observations that will be per-
formed while maximizing their weighted cumulative benefit, and propose a feasi-
ble schedule for these observations. As previous work mostly focused on heuristic
and iterative search algorithms, this paper presents a new technique for selecting
and scheduling observations based on Graph Neural Networks (GNNs) and Deep
Reinforcement Learning (DRL). GNNs are used to extract relevant information
from the graphs representing instances of the EOSP, and DRL drives the search
for optimal schedules. Our simulations show that it is able to learn on small prob-
lem instances and generalize to larger real-world instances, with very competitive
performance compared to traditional approaches. | [
"Scheduling",
"Graph neural Networks",
"Reinforcement learning",
"Earth Observation"
] | https://openreview.net/pdf?id=RlfCQ4cvmd | CpSQE4f1bI | decision | 1,722,287,918,394 | RlfCQ4cvmd | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
RgPrEDav8H | Differentially Private Deep Model-Based Reinforcement Learning | [
"Alexandre Rio",
"Merwan Barlier",
"Igor Colin",
"Albert Thomas"
] | We address deep offline reinforcement learning with privacy guarantees, where the goal is to train a policy that is differentially private with respect to individual trajectories in the dataset. To achieve this, we introduce DP-MORL, an MBRL algorithm with differential privacy guarantees. A private model of the environment is first learned from offline data using DP-FedAvg, a training method for neural networks that provides differential privacy guarantees at the trajectory level. Then, we use model-based policy optimization to derive a policy from the (penalized) private model, without any further interaction with the system or access to the dataset. We empirically show that DP-MORL enables the training of private RL agents from offline data in continuous control tasks and we furthermore outline the price of privacy in this setting. | [
"machine learning",
"reinforcement learning",
"privacy",
"differential privacy",
"deep learning",
"model-based",
"offline"
] | https://openreview.net/pdf?id=RgPrEDav8H | 2Npp1m8KUV | decision | 1,722,287,921,318 | RgPrEDav8H | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Re59h4ew9o | Environment Complexity and Nash Equilibria in a Sequential Social Dilemma | [
"Mustafa Yasir",
"Andrew Howes",
"Vasilios Mavroudis",
"Chris Hicks"
] | Multi-agent reinforcement learning (MARL) methods, while effective in zero-sum or positive-sum games, often yield suboptimal outcomes in general-sum games where cooperation is essential for achieving globally optimal outcomes. Matrix game social dilemmas, which abstract key aspects of general-sum interactions, such as cooperation, risk, and trust, fail to model the temporal and spatial dynamics characteristic of real-world scenarios. In response, our study extends matrix game social dilemmas into more complex, higher-dimensional MARL environments. We adapt a gridworld implementation of the Stag Hunt dilemma to more closely match the decision-space of a one-shot matrix game while also introducing variable environment complexity. Our findings indicate that as complexity increases, MARL agents trained in these environments converge to suboptimal strategies, consistent with the risk-dominant Nash equilibria strategies found in matrix games. Our work highlights the impact of environment complexity on achieving optimal outcomes in higher-dimensional game-theoretic MARL environments. | [
"multi agent RL",
"game theory"
] | https://openreview.net/pdf?id=Re59h4ew9o | 0Yys37NlRl | decision | 1,722,287,920,968 | Re59h4ew9o | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: We strongly encourage the authors to release their code. |
RY8ELv8bVD | Augmented Bayesian Policy Search | [
"Mahdi Kallel",
"Debabrota Basu",
"Riad Akrour",
"Carlo D'Eramo"
] | Deterministic policies are often preferred over stochastic ones when implemented on physical systems. They can prevent erratic and harmful behaviors while being easier to implement and interpret. However, in practice, exploration is largely performed by stochastic policies.
First-order Bayesian Optimization (BO) methods offer a principled way of performing exploration using deterministic policies. This is done through a learned probabilistic model, in the form of a Gaussian Process (GP), of the objective function and its gradient. Nonetheless, such approaches treat policy search as a black-box problem, and thus, neglect the reinforcement learning nature of the problem. In this work, we leverage the performance difference lemma to introduce a novel mean function for the GP. This results in augmenting BO methods with the action-value function. Hence, we call our method Augmented Bayesian Search (ABS). Interestingly, this new mean function enhances the posterior gradient with the deterministic policy gradient, effectively bridging the gap between BO and policy gradient methods. The resulting algorithm combines the convenience of the direct policy search with the scalability of reinforcement learning. We validate ABS on high-dimensional locomotion problems and demonstrate competitive performance compared to existing direct policy search schemes. | [
"Reinforcement learning",
"Policy search",
"Bayesian optimization",
"Gaussian Processes"
] | https://openreview.net/pdf?id=RY8ELv8bVD | WezYorjIAi | decision | 1,722,287,919,963 | RY8ELv8bVD | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept
comment: This contribution appears it could benefit from clarification and better discussion of limitations. We encourage the authors to address the reviewers' comments in their final version and in future submissions. |
RJLSIiNr4D | Adaptive Distributional Double Q-learning | [
"Leif Döring",
"Maximilian Birr",
"Mihail Bîrsan"
] | Bias problems in the estimation of maxima of random variables are a well-known obstacle that drastically slows down $Q$-learning algorithms. We propose to use additional insight gained from distributional reinforcement learning to deal with the overestimation in a locally adaptive way. This helps to combine the strengths and weaknesses of the different $Q$-learning variants in a unified framework. Our framework ADDQ is simple to implement, existing RL algorithms can be improved with a few lines of additional code. We provide experimental results in tabular, Atari, and MuJoCo environments for discrete and continuous control problems, comparisons with state-of-the-art methods, and a proof of convergence. | [
"distributional reinforcement learning",
"double Q learning",
"adaptive learning",
"DQN",
"Actor-Critic"
] | https://openreview.net/pdf?id=RJLSIiNr4D | L0D69nrUiy | decision | 1,722,287,920,199 | RJLSIiNr4D | [
"everyone"
] | [
"EWRL/2024/Workshop/Program_Chairs"
] | title: Paper Decision
decision: Accept |
Subsets and Splits