title
stringlengths 5
246
| categories
stringlengths 5
94
⌀ | abstract
stringlengths 54
5.03k
| authors
stringlengths 0
6.72k
| doi
stringlengths 12
54
⌀ | id
stringlengths 6
10
⌀ | year
float64 2.02k
2.02k
⌀ | venue
stringclasses 13
values |
|---|---|---|---|---|---|---|---|
RISP: Rendering-Invariant State Predictor with Differentiable Simulation and Rendering for Cross-Domain Parameter Estimation | null | This work considers identifying parameters characterizing a physical system's dynamic motion directly from a video whose rendering configurations are inaccessible. Existing solutions require massive training data or lack generalizability to unknown rendering configurations. We propose a novel approach that marries domain randomization and differentiable rendering gradients to address this problem. Our core idea is to train a rendering-invariant state-prediction (RISP) network that transforms image differences into state differences independent of rendering configurations, e.g., lighting, shadows, or material reflectance. To train this predictor, we formulate a new loss on rendering variances using gradients from differentiable rendering. Moreover, we present an efficient, second-order method to compute the gradients of this loss, allowing it to be integrated seamlessly into modern deep learning frameworks. We evaluate our method in rigid-body and deformable-body simulation environments using four tasks: state estimation, system identification, imitation learning, and visuomotor control. We further demonstrate the efficacy of our approach on a real-world example: inferring the state and action sequences of a quadrotor from a video of its motion sequences. Compared with existing methods, our approach achieves significantly lower reconstruction errors and has better generalizability among unknown rendering configurations. | Pingchuan Ma, Tao Du, Joshua B. Tenenbaum, Wojciech Matusik, Chuang Gan | null | null | 2,022 | iclr |
Resolving Training Biases via Influence-based Data Relabeling | null | The performance of supervised learning methods easily suffers from the training bias issue caused by train-test distribution mismatch or label noise. Influence function is a technique that estimates the impacts of a training sample on the model’s predictions. Recent studies on \emph{data resampling} have employed influence functions to identify \emph{harmful} training samples that will degrade model's test performance. They have shown that discarding or downweighting the identified harmful training samples is an effective way to resolve training biases. In this work, we move one step forward and propose an influence-based relabeling framework named RDIA for reusing harmful training samples toward better model performance. To achieve this, we use influence functions to estimate how relabeling a training sample would affect model's test performance and further develop a novel relabeling function R. We theoretically prove that applying R to relabel harmful training samples allows the model to achieve lower test loss than simply discarding them for any classification tasks using cross-entropy loss. Extensive experiments on ten real-world datasets demonstrate RDIA outperforms the state-of-the-art data resampling methods and improves model's robustness against label noise. | Shuming Kong, Yanyan Shen, Linpeng Huang | null | null | 2,022 | iclr |
Diffusion-Based Voice Conversion with Fast Maximum Likelihood Sampling Scheme | null | Voice conversion is a common speech synthesis task which can be solved in different ways depending on a particular real-world scenario. The most challenging one often referred to as one-shot many-to-many voice conversion consists in copying target voice from only one reference utterance in the most general case when both source and target speakers do not belong to the training dataset. We present a scalable high-quality solution based on diffusion probabilistic modeling and demonstrate its superior quality compared to state-of-the-art one-shot voice conversion approaches. Moreover, focusing on real-time applications, we investigate general principles which can make diffusion models faster while keeping synthesis quality at a high level. As a result, we develop a novel Stochastic Differential Equations solver suitable for various diffusion model types and generative tasks as shown through empirical studies and justify it by theoretical analysis. | Vadim Popov, Ivan Vovk, Vladimir Gogoryan, Tasnima Sadekova, Mikhail Sergeevich Kudinov, Jiansheng Wei | null | null | 2,022 | iclr |
CycleMLP: A MLP-like Architecture for Dense Prediction | null | This paper presents a simple MLP-like architecture, CycleMLP, which is a versatile backbone for visual recognition and dense predictions. As compared to modern MLP architectures, e.g. , MLP-Mixer, ResMLP, and gMLP, whose architectures are correlated to image size and thus are infeasible in object detection and segmentation, CycleMLP has two advantages compared to modern approaches. (1) It can cope
with various image sizes. (2) It achieves linear computational complexity to image size by using local windows. In contrast, previous MLPs have $O(N^2)$ computations due to fully spatial connections. We build a family of models which surpass existing MLPs and even state-of-the-art Transformer-based models, e.g. Swin Transformer, while using fewer parameters and FLOPs. We expand the MLP-like models’ applicability, making them a versatile backbone for dense prediction tasks. CycleMLP achieves competitive results on object detection, instance segmentation, and semantic segmentation. In particular, CycleMLP-Tiny outperforms Swin-Tiny by 1.3% mIoU on ADE20K dataset with fewer FLOPs. Moreover, CycleMLP also shows excellent zero-shot robustness on ImageNet-C dataset. | Shoufa Chen, Enze Xie, Chongjian GE, Runjian Chen, Ding Liang, Ping Luo | null | null | 2,022 | iclr |
Fairness-aware Federated Learning | null | Federated Learning is a machine learning technique where a network of clients collaborates with a server to learn a centralized model while keeping data localized. In such a setting, naively minimizing an aggregate loss may introduce bias and disadvantage model performance on certain clients. To address this issue, we propose a new federated learning framework called FAFL in which the goal is to minimize the worst-case weighted client losses over an uncertainty set. By deriving a variational representation, we show that this framework is a fairness-aware objective and can be easily optimized by solving a joint minimization problem over the model parameters and a dual variable. We then propose an optimization algorithm to solve FAFL which can be efficiently implemented in a federated setting and provide convergence guarantees. We further prove generalization bounds for learning with this objective. Experiments on real-world datasets demonstrate the effectiveness of our framework in achieving both accuracy and fairness. | Zhuozhuo Tu, zhiqiang xu, Tairan Huang, Dacheng Tao, Ping Li | null | null | 2,022 | iclr |
On the Implicit Biases of Architecture & Gradient Descent | null | Do neural networks generalise because of bias in the functions returned by gradient descent, or bias already present in the network architecture? $\textit{¿Por qué no los dos?}$ This paper finds that while typical networks that fit the training data already generalise fairly well, gradient descent can further improve generalisation by selecting networks with a large margin. This conclusion is based on a careful study of the behaviour of infinite width networks trained by Bayesian inference and finite width networks trained by gradient descent. To measure the implicit bias of architecture, new technical tools are developed to both $\textit{analytically bound}$ and $\textit{consistently estimate}$ the average test error of the neural network--Gaussian process (NNGP) posterior. This error is found to be already better than chance, corroborating the findings of Valle-Pérez et al. (2019) and underscoring the importance of architecture. Going beyond this result, this paper finds that test performance can be substantially improved by selecting a function with much larger margin than is typical under the NNGP posterior. This highlights a curious fact: $\textit{minimum a posteriori}$ functions can generalise best, and gradient descent can select for those functions. In summary, new technical tools suggest a nuanced portrait of generalisation involving both the implicit biases of architecture and gradient descent. | Jeremy Bernstein, Yisong Yue | null | null | 2,022 | iclr |
Q-Learning Scheduler for Multi-Task Learning through the use of Histogram of Task Uncertainty | null | Simultaneous training of a multi-task learning network on different domains or tasks is not always straightforward. It could lead to inferior performance or generalization compared to the corresponding single-task networks. To maximally taking advantage of the benefits of multi-task learning, an effective training scheduling method is deemed necessary. Traditional schedulers follow a heuristic or prefixed strategy, ignoring the relation of the tasks, their sample complexities, and the state of the emergent shared features. We proposed a deep Q-Learning Scheduler (QLS) that monitors the state of the tasks and the shared features using a novel histogram of task uncertainty, and through trial-and-error, learns an optimal policy for task scheduling. Extensive experiments on multi-domain and multi-task settings with various task difficulty profiles have been conducted, the proposed method is benchmarked against other schedulers, its superior performance has been demonstrated, and results are discussed. | Kourosh Meshgi, Maryam Sadat Mirzaei, Satoshi Sekine | null | null | 2,022 | iclr |
A new perspective on probabilistic image modeling | null | We present the Deep Convolutional Gaussian Mixture Model (DCGMM), a new probabilistic approach for image modeling capable of density estimation, sampling and tractable inference. DCGMM instances exhibit a CNN-like layered structure, in which the principal building blocks are convolutional Gaussian Mixture (cGMM) layers. A key innovation w.r.t. related models lile sum-produdct networks (SPNs) and probabilistic circuits (PCs) is that each cGMM layer optimizes an independent loss function and therefore has an independent probabilistic interpretation. This modular approach permits intervening transformation layers to harness the full spectrum of
(potentially non-invertible) mappings available to CNNs, e.g., max-pooling or (half-)convolutions. DCGMM sampling and inference are realized by a deep chain of hierarchical priors, where samples generated by each cGMM layer parameterize sampling in the next-lower cGMM layer. For sampling through non-invertible transformation layers, we introduce a new gradient-based sharpening technique that exploits redundancy (overlap) in, e.g., half-convolutions. The basic quantities forward-transported through a DCGMM instance are the posterior probabilities of cGMM layers, which ensures numerical stability and facilitates the selection of learning rates.
DCGMMs can be trained end-to-end by SGD from random initial conditions, much like CNNs. We experimentally show that DCGMMs compare favorably to several recent PC and SPN models in terms of inference, classification and sampling, the latter particularly for challenging datasets such as SVHN. A public TF2 implementation is provided as well. | Alexander Gepperth | null | null | 2,022 | iclr |
Provable Identifiability of ReLU Neural Networks via Lasso Regularization | null | LASSO regularization is a popular regression tool to enhance the prediction accuracy of statistical models by performing variable selection through the $\ell_1$ penalty, initially formulated for the linear model and its variants. In this paper, the territory of LASSO is extended to the neural network model, a fashionable and powerful nonlinear regression model. Specifically, given a neural network whose output $y$ depends only on a small subset of input $\boldsymbol{x}$, denoted by $\mathcal{S}^{\star}$, we prove that the LASSO estimator can stably reconstruct the neural network and identify $\mathcal{S}^{\star}$ when the number of samples scales logarithmically with the input dimension. This challenging regime has been well understood for linear models while barely studied for neural networks. Our theory lies in an extended Restricted Isometry Property (RIP)-based analysis framework for two-layer ReLU neural networks, which may be of independent interest to other LASSO or neural network settings. Based on the result, we further propose a neural network-based variable selection method. Experiments on simulated and real-world datasets show the promising performance of our variable selection approach compared with classical techniques. | Gen Li, Ganghua Wang, Yuantao Gu, Jie Ding | null | null | 2,022 | iclr |
SpanDrop: Simple and Effective Counterfactual Learning for Long Sequences | null | Distilling supervision signal from a long sequence to make predictions is a challenging task in machine learning, especially when not all elements in the input sequence contribute equally to the desired output. In this paper, we propose SpanDrop, a simple and effective data augmentation technique that helps models identify the true supervision signal in a long sequence with very few examples. By directly manipulating the input sequence, SpanDrop randomly ablates parts of the sequence at a time and ask the model to perform the same task to emulate counterfactual learning and achieve input attribution. Based on theoretical analysis of its properties, we also propose a variant of SpanDrop based on the beta-Bernoulli distribution, which yields diverse augmented sequences while providing a learning objective that is more consistent with the original dataset. We demonstrate the effectiveness of SpanDrop on a set of carefully designed toy tasks, as well as various natural language processing tasks that require reasoning over long sequences to arrive at the correct answer, and show that it helps models improve performance both when data is scarce and abundant. | Peng Qi, Guangtao Wang, Jing Huang | null | null | 2,022 | iclr |
On the Uncomputability of Partition Functions in Energy-Based Sequence Models | null | In this paper, we argue that energy-based sequence models backed by expressive parametric families can result in uncomputable and inapproximable partition functions. Among other things, this makes model selection--and therefore learning model parameters--not only difficult, but generally _undecidable_. The reason is that there are no good deterministic or randomized estimates of partition functions. Specifically, we exhibit a pathological example where under common assumptions, _no_ useful importance sampling estimates of the partition function can guarantee to have variance bounded below a rational number. As alternatives, we consider sequence model families whose partition functions are computable (if they exist), but at the cost of reduced expressiveness. Our theoretical results suggest that statistical procedures with asymptotic guarantees and sheer (but finite) amounts of compute are not the only things that make sequence modeling work; computability concerns must not be neglected as we consider more expressive model parametrizations. | Chu-Cheng Lin, Arya D. McCarthy | null | null | 2,022 | iclr |
PulseImpute: A Novel Benchmark Task and Architecture for Imputation of Physiological Signals | null | Providing care for patients with chronic diseases is one of the biggest drivers of the nation’s rising healthcare costs, but many of these diseases are linked to mutable health behaviors. Mobile health (mHealth) biophysical sensors that continuously measure our current conditions provide the framework for a personalized guidance system for the maintenance of healthy behaviors. However, this physiological sensor data is plagued with missingness due to insecure attachments, wireless dropout, battery, and adherence issues. These issues cripple their rich diagnostic utility as well as their ability to enable temporally-precise interventions. While there is a sizable amount of research focusing on imputation methods, surprisingly, no works have addressed the patterns of missingness, quasi-periodic signal structure, and the between subject heterogeneity that characterizes physiological signals in mHealth applications. We present the PulseImpute Challenge, the first challenge dataset for physiological signal imputation which includes a large set of baselines' performances on realistic missingness models and data. Next, we demonstrate the potential to address this quasi-periodic structure and heterogeneity with our Dilated Convolution Bottleneck (DCB) Transformer, a transformer architecture with a self-attention mechanism that is able to attend to corresponding waveform features in quasi-periodic signals. By utilizing stacked dilated convolutions with bottleneck layers for query and key transformations, we visually demonstrate that the kernel similarity in the attention model gives high similarity to similar temporal features across quasi-periodic periods. We hope the release of our challenge task definitions and baseline implementations will spur the community to address this challenging and important problem.
| Maxwell Xu, Alexander Moreno, James Matthew Rehg | null | null | 2,022 | iclr |
DEGREE: Decomposition Based Explanation for Graph Neural Networks | null | Graph Neural Networks (GNNs) are gaining extensive attention for their application in graph data. However, the black-box nature of GNNs prevents users from understanding and trusting the models, thus hampering their applicability. Whereas explaining GNNs remains a challenge, most existing methods fall into approximation based and perturbation based approaches with suffer from faithfulness problems and unnatural artifacts respectively. To tackle these problems, we propose DEGREE (Decomposition based Explanation for GRaph nEural nEtworks) to provide a faithful explanation for GNN predictions. By decomposing the information generation and aggregation mechanism of GNNs, DEGREE allows tracking the contributions of specific components of the input graph to the final prediction. Based on this, we further design a subgraph level interpretation algorithm to reveal complex interactions between graph nodes that are overlooked by previous methods. The efficiency of our algorithm can be further improved by utilizing GNN characteristics. Finally, we conduct quantitative and qualitative experiments on synthetic and real-world datasets to demonstrate the effectiveness of DEGREE on node classification and graph classification tasks. | Qizhang Feng, Ninghao Liu, Fan Yang, Ruixiang Tang, Mengnan Du, Xia Hu | null | null | 2,022 | iclr |
Evaluating Distributional Distortion in Neural Language Modeling | null | A fundamental characteristic of natural language is the high rate at which speakers produce novel expressions. Because of this novelty, a heavy-tail of rare events accounts for a significant amount of the total probability mass of distributions in language (Baayen, 2001). Standard language modeling metrics such as perplexity quantify the performance of language models (LM) in aggregate. As a result, we have relatively little understanding of whether neural LMs accurately estimate the probability of sequences in this heavy-tail of rare events. To address this gap, we develop a controlled evaluation scheme which uses generative models trained on natural data as artificial languages from which we can exactly compute sequence probabilities. Training LMs on generations from these artificial languages, we compare the sequence-level probability estimates given by LMs to the true probabilities in the target language. Our experiments reveal that LSTM and Transformer language models (i) systematically underestimate the probability of sequences drawn from the target language, and (ii) do so more severely for less-probable sequences. Investigating where this probability mass went, (iii) we find that LMs tend to overestimate the probability of ill formed (perturbed) sequences. In addition, we find that this underestimation behaviour (iv) is weakened, but not eliminated by greater amounts of training data, and (v) is exacerbated for target distributions with lower entropy. | Benjamin LeBrun, Alessandro Sordoni, Timothy J. O'Donnell | null | null | 2,022 | iclr |
KINet: Keypoint Interaction Networks for Unsupervised Forward Modeling | null | Object-centric representation is an essential abstraction for physical reasoning and forward prediction. Most existing approaches learn this representation through extensive supervision (e.g, object class and bounding box) although such ground-truth information is not readily accessible in reality. To address this, we introduce KINet (Keypoint Interaction Network)---an end-to-end unsupervised framework to reason about object interactions in complex systems based on a keypoint representation. Using visual observations, our model learns to associate objects with keypoint coordinates and discovers a graph representation of the system as a set of keypoint embeddings and their relations. It then learns an action-conditioned forward model using contrastive estimation to predict future keypoint states. By learning to perform physical reasoning in the keypoint space, our model automatically generalizes to scenarios with a different number of objects, and novel object geometries. Experiments demonstrate the effectiveness of our model to accurately perform forward prediction and learn plannable object-centric representations which can also be used in downstream model-based control tasks. | Alireza Rezazadeh, Changhyun Choi | null | null | 2,022 | iclr |
Increase and Conquer: Training Graph Neural Networks on Growing Graphs | null | Graph neural networks (GNNs) use graph convolutions to exploit network invariances and learn meaningful features from network data. However, on large-scale graphs convolutions incur in high computational cost, leading to scalability limitations. Leveraging the graphon --- the limit object of a graph --- in this paper we consider the problem of learning a graphon neural network (WNN) --- the limit object of a GNN --- by training GNNs on graphs sampled Bernoulli from the graphon. Under smoothness conditions, we show that: (i) the expected distance between the learning steps on the GNN and on the WNN decreases asymptotically with the size of the graph, and (ii) when training on a sequence of growing graphs, gradient descent follows the learning direction of the WNN. Inspired by these results, we propose a novel algorithm to learn GNNs on large-scale graphs that, starting from a moderate number of nodes, successively increases the size of the graph during training. This algorithm is benchmarked on both a recommendation system and a decentralized control problem where it is shown to retain comparable performance, to its large-scale counterpart, at a reduced computational cost. | Juan Cervino, Luana Ruiz, Alejandro Ribeiro | null | null | 2,022 | iclr |
SSR-GNNs: Stroke-based Sketch Representation with Graph Neural Networks | null | Existing end-to-end visual recognition models do not possess innate spatial invariance and are thus vulnerable to out-of-training attacks. This suggests the need of a better representation design. This paper follows existing cognitive studies to investigate a sketch representation that specify stroke information on vertices and inter-stroke information on edges. The resultant representation, combined with a graph neural network, achieves both high classification accuracy and high robustness against translation, rotation, and stroke-wise parametric and topological attacks thanks to the use of spatially invariant stroke features and GNN architecture. While prior studies demonstrated similar sketch representations for classification and generation, these attempts heavily relied on run-time statistical inference rather than more efficient bottom-up computation via GNN. The presented sketch representation poses good structured expression capability as it enables generation of sketches semantically different from the training dataset. Lastly, we show SSR-GNNs are able to accomplish all tasks (classification, robust feature learning, and novel pattern generation), which shows that the representation is task-agnostic. | Sheng Cheng, Yi Ren, Yezhou Yang | null | null | 2,022 | iclr |
E-LANG: Energy-based Joint Inferencing of Super and Swift Language Models | null | Building very large and highly capable language models has been a trend in the past several years. Despite their great performance, they incur a high computational cost. A common solution is to apply model compression or choose light-weight architectures, which often need a separate fixed-size model for each desirable computational budget, and may lose performance in case of heavy compression. This paper proposes an effective dynamic inference approach, which distributes the inference between large accurate Super-models and light-weight Swift models. To this end, a decision making module routes the incoming samples to one of the two models based on the energy characteristics of the representations in the latent space. The proposed approach is easily adoptable and architecture agnostic. As such, it can be applied to black-box pre-trained models without a need for architectural manipulations, careful reassembling of modules, or re-training. Unlike existing methods that are for the most part only applicable to encoder-only backbones and classification tasks, our method also works for encoder-decoder structures and sequence-to-sequence tasks such as translation. The performance of the proposed Energy-based joint inferencing of LANGuage models, E-LANG, is verified through an extensive set of experiments with T5 and BERT architectures on GLUE, SuperGLUE, and WMT benchmarks. In particular, we outperform T5-11B with an average computations speed-up of 3.3X on GLUE and 2.9X on SuperGLUE. We also achieve BERT-based SOTA (state-of-the-art) on GLUE with 3.2X less computations. Code is available in the supplementary materials. | Mohammad Akbari, Amin Banitalebi-Dehkordi, Yong Zhang | null | null | 2,022 | iclr |
Equivariant Grasp learning In Real Time | null | Visual grasp detection is a key problem in robotics where the agent must learn to model the grasp function, a mapping from an image of a scene onto a set of feasible grasp poses. In this paper, we recognize that the grasp function is $\mathrm{SE}(2)$-equivariant and that it can be modeled using an equivariant convolutional neural network. As a result, we are able to significantly improve the sample efficiency of grasp learning to the point where we can learn a good approximation of the grasp function within only 500 grasp experiences. This is fast enough that we can learn to grasp completely on a physical robot in about an hour. | Xupeng Zhu, Dian Wang, Ondrej Biza, Robert Platt | null | null | 2,022 | iclr |
Inductive Bias of Multi-Channel Linear Convolutional Networks with Bounded Weight Norm | null | We provide a function space characterization of the inductive bias resulting from minimizing the $\ell_2$ norm of the weights in multi-channel linear convolutional networks. We define an \textit{induced regularizer} in the function space as the minimum $\ell_2$ norm of weights of a network required to realize a function. For two layer linear convolutional networks with $C$ output channels and kernel size $K$, we show the following: (a) If the inputs to the network have a single channel, the induced regularizer for any $K$ is \textit{independent} of the number of output channels $C$. Furthermore, we derive the regularizer is a norm given by a semidefinite program (SDP). (b) In contrast, for networks with multi-channel inputs, multiple output channels can be necessary to merely realize all matrix-valued linear functions and thus the inductive bias \emph{does} depend on $C$. However, for sufficiently large $C$, the induced regularizer is again given by an SDP that is independent of $C$. In particular, the induced regularizer for $K=1$ and $K=D$ are given in closed form as the nuclear norm and the $\ell_{2,1}$ group-sparse norm, respectively, of the Fourier coefficients.
We investigate the applicability of our theoretical results to a broader scope of ReLU convolutional networks through experiments on MNIST and CIFAR-10 datasets. | Meena Jagadeesan, Ilya Razenshteyn, Suriya Gunasekar | null | null | 2,022 | iclr |
Ask2Mask: Guided Data Selection for Masked Speech Modeling | null | Masked speech modeling (MSM) methods such as wav2vec2 or w2v-BERT learn representations over speech frames which are randomly masked within an utterance. While these methods improve performance of Automatic Speech Recognition (ASR) systems, they have one major limitation. They treat all unsupervised speech samples with equal weight, which hinders learning as not all samples have relevant information to learn meaningful representations. In this work, we address this limitation. We propose ask2mask (ATM), a novel approach to focus on specific samples during MSM pre-training. ATM employs an external ASR model or \textit{scorer} to weight unsupervised input samples in two different ways: 1) A fine-grained data selection is performed by masking over the highly confident input frames as chosen by the scorer. This allows the model to learn meaningful representations. 2) ATM is further extended to focus at utterance-level by weighting the final MSM loss with the utterance-level confidence score. We conduct fine-tuning experiments on two well-benchmarked corpora: LibriSpeech (matching the pre-training data) and AMI (not matching the pre-training data). The results substantiate the efficacy of ATM on significantly improving the recognition performance under mismatched conditions (up to 11.6\% relative) while still yielding modest improvements under matched conditions. | Murali Karthick Baskar, Andrew Rosenberg, Bhuvana Ramabhadran, Yu Zhang, Pedro Moreno | null | null | 2,022 | iclr |
GeoDiff: A Geometric Diffusion Model for Molecular Conformation Generation | null | Predicting molecular conformations from molecular graphs is a fundamental problem in cheminformatics and drug discovery. Recently, significant progress has been achieved with machine learning approaches, especially with deep generative models. Inspired by the diffusion process in classical non-equilibrium thermodynamics where heated particles will diffuse from original states to a noise distribution, in this paper, we propose a novel generative model named GeoDiff for molecular conformation prediction. GeoDiff treats each atom as a particle and learns to directly reverse the diffusion process (i.e., transforming from a noise distribution to stable conformations) as a Markov chain. Modeling such a generation process is however very challenging as the likelihood of conformations should be roto-translational invariant. We theoretically show that Markov chains evolving with equivariant Markov kernels can induce an invariant distribution by design, and further propose building blocks for the Markov kernels to preserve the desirable equivariance property. The whole framework can be efficiently trained in an end-to-end fashion by optimizing a weighted variational lower bound to the (conditional) likelihood. Experiments on multiple benchmarks show that GeoDiff is superior or comparable to existing state-of-the-art approaches, especially on large molecules. | Minkai Xu, Lantao Yu, Yang Song, Chence Shi, Stefano Ermon, Jian Tang | null | null | 2,022 | iclr |
PEARL: Data Synthesis via Private Embeddings and Adversarial Reconstruction Learning | null | We propose a new framework of synthesizing data using deep generative models in a differentially private manner.
Within our framework, sensitive data are sanitized with rigorous privacy guarantees in a one-shot fashion, such that training deep generative models is possible without re-using the original data.
Hence, no extra privacy costs or model constraints are incurred, in contrast to popular gradient sanitization approaches, which, among other issues, cause degradation in privacy guarantees as the training iteration increases.
We demonstrate a realization of our framework by making use of the characteristic function and an adversarial re-weighting objective, which are of independent interest as well.
Our proposal has theoretical guarantees of performance, and empirical evaluations on multiple datasets show that our approach outperforms other methods at reasonable levels of privacy. | Seng Pei Liew, Tsubasa Takahashi, Michihiko Ueno | null | null | 2,022 | iclr |
Linking Emergent and Natural Languages via Corpus Transfer | null | The study of language emergence aims to understand how human languages are shaped by perceptual grounding and communicative intent. Computational approaches to emergent communication (EC) predominantly consider referential games in limited domains and analyze the learned protocol within the game framework. As a result, it remains unclear how the emergent languages from these settings connect to natural languages or provide benefits in real-world language processing tasks, where statistical models trained on large text corpora dominate. In this work, we propose a novel way to establish such a link by corpus transfer, i.e. pretraining on a corpus of emergent language for downstream natural language tasks, which is in contrast to prior work that directly transfers speaker and listener parameters. Our approach showcases non-trivial transfer benefits for two different tasks – language modeling and image captioning. For example, in a low-resource setup (modeling 2 million natural language tokens), pre-training on an emergent language corpus with just 2 million tokens reduces model perplexity by 24.6% on average across ten natural languages. We also introduce a novel metric to predict the transferability of an emergent language by translating emergent messages to natural language captions grounded on the same images. We find that our translation-based metric highly correlates with the downstream performance on modeling natural languages (for instance $\rho = 0.83$ on Hebrew), while topographic similarity, a popular metric in previous works, shows surprisingly low correlation ($\rho = 0.003$), hinting that simple properties like attribute disentanglement from synthetic domains might not capture the full complexities of natural language. Our findings also indicate potential benefits of moving language emergence forward with natural language resources and models. | Shunyu Yao, Mo Yu, Yang Zhang, Karthik R Narasimhan, Joshua B. Tenenbaum, Chuang Gan | null | null | 2,022 | iclr |
GreaseLM: Graph REASoning Enhanced Language Models | null | Answering complex questions about textual narratives requires reasoning over both stated context and the world knowledge that underlies it. However, pretrained language models (LM), the foundation of most modern QA systems, do not robustly represent latent relationships between concepts, which is necessary for reasoning. While knowledge graphs (KG) are often used to augment LMs with structured representations of world knowledge, it remains an open question how to effectively fuse and reason over the KG representations and the language context, which provides situational constraints and nuances. In this work, we propose GreaseLM, a new model that fuses encoded representations from pretrained LMs and graph neural networks over multiple layers of modality interaction operations. Information from both modalities propagates to the other, allowing language context representations to be grounded by structured world knowledge, and allowing linguistic nuances (e.g., negation, hedging) in the context to inform the graph representations of knowledge. Our results on three benchmarks in the commonsense reasoning (i.e., CommonsenseQA, OpenbookQA) and medical question answering (i.e., MedQA-USMLE) domains demonstrate that GreaseLM can more reliably answer questions that require reasoning over both situational constraints and structured knowledge, even outperforming models 8x larger. | Xikun Zhang, Antoine Bosselut, Michihiro Yasunaga, Hongyu Ren, Percy Liang, Christopher D Manning, Jure Leskovec | null | null | 2,022 | iclr |
DISCOVERING AND EXPLAINING THE REPRESENTATION BOTTLENECK OF DNNS | null | This paper explores the bottleneck of feature representations of deep neural networks (DNNs), from the perspective of the complexity of interactions between input variables encoded in DNNs. To this end, we focus on the multi-order interaction between input variables, where the order represents the complexity of interactions. We discover that a DNN is more likely to encode both too simple and too complex interactions, but usually fails to learn interactions of intermediate complexity. Such a phenomenon is widely shared by different DNNs for different tasks. This phenomenon indicates a cognition gap between DNNs and humans, and we call it a representation bottleneck. We theoretically prove the underlying reason for the representation bottleneck. Furthermore, we propose losses to encourage/penalize the learning of interactions of specific complexities, and analyze the representation capacities of interactions of different complexities. The code is available at https://github.com/Nebularaid2000/bottleneck. | Huiqi Deng, Qihan Ren, Hao Zhang, Quanshi Zhang | null | null | 2,022 | iclr |
Analytic-DPM: an Analytic Estimate of the Optimal Reverse Variance in Diffusion Probabilistic Models | null | Diffusion probabilistic models (DPMs) represent a class of powerful generative models. Despite their success, the inference of DPMs is expensive since it generally needs to iterate over thousands of timesteps. A key problem in the inference is to estimate the variance in each timestep of the reverse process. In this work, we present a surprising result that both the optimal reverse variance and the corresponding optimal KL divergence of a DPM have analytic forms w.r.t. its score function. Building upon it, we propose \textit{Analytic-DPM}, a training-free inference framework that estimates the analytic forms of the variance and KL divergence using the Monte Carlo method and a pretrained score-based model. Further, to correct the potential bias caused by the score-based model, we derive both lower and upper bounds of the optimal variance and clip the estimate for a better result. Empirically, our analytic-DPM improves the log-likelihood of various DPMs, produces high-quality samples, and meanwhile enjoys a $20\times$ to $80\times$ speed up. | Fan Bao, Chongxuan Li, Jun Zhu, Bo Zhang | null | null | 2,022 | iclr |
Semi-supervised Offline Reinforcement Learning with Pre-trained Decision Transformers | null | Pre-training deep neural network models using large unlabelled datasets followed by fine-tuning them on small task-specific datasets has emerged as a dominant paradigm in natural language processing (NLP) and computer vision (CV). Despite the widespread success, such a paradigm has remained atypical in reinforcement learning (RL).
In this paper, we investigate how we can leverage large reward-free (i.e. task-agnostic) offline datasets of prior interactions to pre-train agents that can then be fine-tuned using a small reward-annotated dataset. To this end, we present Pre-trained Decision Transformer (PDT), a simple yet powerful algorithm for semi-supervised Offline RL. By masking reward tokens during pre-training, the transformer learns to autoregressivley predict actions based on previous state and action context and effectively extracts behaviors present in the dataset. During fine-tuning, rewards are un-masked and the agent learns the set of skills that should be invoked for the desired behavior as per the reward function. We demonstrate the efficacy of this simple and flexible approach on tasks from the D4RL benchmark with limited reward annotations. | Catherine Cang, Kourosh Hakhamaneshi, Ryan Rudes, Igor Mordatch, Aravind Rajeswaran, Pieter Abbeel, Michael Laskin | null | null | 2,022 | iclr |
Generate Triggers in Neural Relation Extraction | null | In the relation extraction task, the relationship between two entities is determined by some specific words in their source text. These words are called relation triggers, which are the evidence to explain the relationship; other words are called ir-relevant words. The current relationship extraction neural network model aims at identifying the relation type between two entities mentioned in source text by encoding the text and entities. However, these models cannot output the relation triggers, but only gives the result of relation classification. Although models can generate weights for every single word through the improvement of attention mechanism, the weights will be affected by irrelevant words essentially, which are not required by the relation extraction task. In order to output re-lation triggers accurately, we propose a novel training frame-work for Relation Extraction (RE) that reduces the negative effect of irrelevant words on them in the encoding stage. In specific, we leverage Evolutive Mask based Point Network (EMPN) as a decoder to generate relation triggers and encode these words again. For an ordered output in relation triggers, we utilize order loss to constrain the output order in them. Ex-tensive experiment results demonstrate that the effectiveness of our proposed model achieves state-of-the-art performance on three RE benchmark datasets. | Liu Yujiang | null | null | 2,022 | iclr |
Automatic Integration for Neural Temporal Point Process | null | Integration lies at the heart of the temporal point process. Due to the intrinsic mathematical difficulty of symbolic integration, neural temporal point process models either constrain the intensity function to an integrable functional form or apply certain numerical methods. However, the former type of model has limited expressive power, and the latter type of model suffers additional numerical errors and high computational costs. In this paper, we introduce automatic integration with neural point process models, a new paradigm for efficient, closed-form nonparametric inference of temporal point process characterized by any intensity function. We test the model against a variety of synthetic temporal point process datasets and show that the model can better capture inter-event intensity changes than state-of-the-art methods. We also identify certain model settings that would lead the MLE estimator for the temporal point process to be inconsistent. | Zihao Zhou, Rose Yu | null | null | 2,022 | iclr |
Pretraining Text Encoders with Adversarial Mixture of Training Signal Generators | null | We present a new framework AMOS that pretrains text encoders with an Adversarial learning curriculum via a Mixture Of Signals from multiple auxiliary generators. Following ELECTRA-style pretraining, the main encoder is trained as a discriminator to detect replaced tokens generated by auxiliary masked language models (MLMs). Different from ELECTRA which trains one MLM as the generator, we jointly train multiple MLMs of different sizes to provide training signals at various levels of difficulty. To push the discriminator to learn better with challenging replaced tokens, we learn mixture weights over the auxiliary MLMs' outputs to maximize the discriminator loss by backpropagating the gradient from the discriminator via Gumbel-Softmax. For better pretraining efficiency, we propose a way to assemble multiple MLMs into one unified auxiliary model. AMOS outperforms ELECTRA and recent state-of-the-art pretrained models by about 1 point on the GLUE benchmark for BERT base-sized models. | Yu Meng, Chenyan Xiong, Payal Bajaj, saurabh tiwary, Paul N. Bennett, Jiawei Han, Xia Song | null | null | 2,022 | iclr |
Brain insights improve RNNs' accuracy and robustness for hierarchical control of continually learned autonomous motor motifs | null | We study the problem of learning dynamics that can produce hierarchically organized continuous outputs consisting of the flexible chaining of re-usable motor ‘motifs’ from which complex behavior is generated. Can a motif library be efficiently and extendably learned without interference between motifs, and can these motifs be chained in arbitrary orders without first learning the corresponding motif transitions during training? This requires (i) parameter updates while learning a new motif that do not interfere with the parameters used for the previously acquired ones; and (ii) successful motif generation when starting from the network states reached at the end of any of the other motifs, even if these states were not present during training (a case of out-of-distribution generalization). We meet the first requirement by designing recurrent neural networks (RNNs) with specific architectures that segregate motif-dependent parameters (as customary in continual learning works), and try a standard method to address the second by training with random initial states. We find that these standard RNNs are very unreliable during zero-shot transfer to motif chaining. We then use insights from the motor thalamocortical circuit, featuring a specific module that shapes motif transitions. We develop a method to constrain the RNNs to function similarly to the thalamocortical circuit during motif transitions, while preserving the large expressivity afforded by gradient-based training of non-analytically tractable RNNs. We then show that this thalamocortical inductive bias not only acts in synergy with gradient-descent RNN training to improve accuracy during in-training-distribution motif production, but also leads to zero-shot transfer to new motif chains with no performance cost. Besides proposing an efficient, robust and flexible RNN architecture, our results shed new light on the function of motor preparation in the brain. | Laureline Logiaco, G Sean Escola | null | null | 2,022 | iclr |
Fast Convergence of Optimistic Gradient Ascent in Network Zero-Sum Extensive Form Games | null | The study of learning in games has thus far focused primarily on normal form games. In contrast, our understanding of learning in extensive form games (EFG) and particularly in EFGs with many agents lags far behind, despite them being closer in nature to many real world applications. We consider the natural class of Network Zero-Sum Extensive Form Games, which combines the global zero-sum property of agent payoffs, the efficient representation of graphical games as well the expressive power of EFGs. We examine the convergence properties of Optimistic Gradient Ascent (OGA) in these games. We prove that the time-average behavior of such online learning dynamics exhibits $O(1/T)$ rate of convergence to the set of Nash equilibria. Moreover, we show that the day-to-day behavior also converges to Nash with rate $O(c^{-t})$ for some game-dependent constant $c > 0$. | Ryann Sim, EFSTRATIOS PANTELEIMON SKOULAKIS, Lillian J Ratliff, Georgios Piliouras | null | null | 2,022 | iclr |
Can an Image Classifier Suffice For Action Recognition? | null | We explore a new perspective on video understanding by casting the video recognition problem as an image recognition task. Our approach rearranges input video frames into super images, which allow for training an image classifier directly to fulfill the task of action recognition, in exactly the same way as image classification. With such a simple idea, we show that transformer-based image classifiers alone can suffice for action recognition. In particular, our approach demonstrates strong and promising performance against SOTA methods on several public datasets including Kinetics400, Moments In Time, Something-Something V2 (SSV2), Jester and Diving48. We also experiment with the prevalent ResNet image classifiers in computer vision to further validate our idea. The results on both Kinetics400 and SSV2 are comparable to some of the best-performed CNN approaches based on spatio-temporal modeling. Our source codes and models are available at \url{https://github.com/IBM/sifar-pytorch}. | Quanfu Fan, Chun-Fu Chen, Rameswar Panda | null | null | 2,022 | iclr |
PiCO: Contrastive Label Disambiguation for Partial Label Learning | null | Partial label learning (PLL) is an important problem that allows each training example to be labeled with a coarse candidate set, which well suits many real-world data annotation scenarios with label ambiguity. Despite the promise, the performance of PLL often lags behind the supervised counterpart. In this work, we bridge the gap by addressing two key research challenges in PLL---representation learning and label disambiguation---in one coherent framework. Specifically, our proposed framework PiCO consists of a contrastive learning module along with a novel class prototype-based label disambiguation algorithm. PiCO produces closely aligned representations for examples from the same classes and facilitates label disambiguation. Theoretically, we show that these two components are mutually beneficial, and can be rigorously justified from an expectation-maximization (EM) algorithm perspective. Extensive experiments demonstrate that PiCO significantly outperforms the current state-of-the-art approaches in PLL and even achieves comparable results to fully supervised learning. Code and data available: https://github.com/hbzju/PiCO. | Haobo Wang, Ruixuan Xiao, Yixuan Li, Lei Feng, Gang Niu, Gang Chen, Junbo Zhao | null | null | 2,022 | iclr |
Memorizing Transformers | null | Language models typically need to be trained or finetuned in order to acquire new knowledge, which involves updating their weights.
We instead envision language models that can simply read and memorize new data at inference time, thus acquiring new knowledge immediately. In this work, we extend language models with the ability to memorize the internal representations of past inputs. We demonstrate that an approximate $k$NN lookup into a non-differentiable memory of recent (key, value) pairs improves language modeling across various benchmarks and tasks, including generic webtext (C4), math papers (arXiv), books (PG-19), code (Github), as well as formal theorems (Isabelle). We show that the performance steadily improves when we increase the size of memory up to 262K tokens.
On benchmarks including code and mathematics, we find that the model is capable of making use of newly defined functions and theorems during test time. | Yuhuai Wu, Markus Norman Rabe, DeLesley Hutchins, Christian Szegedy | null | null | 2,022 | iclr |
Selective Token Generation for Few-shot Language Modeling | null | Natural language modeling with limited training data is challenging problem, and many algorithms make use of large-scale pretrained language models (PLMs) for this due to its great generalization ability. Among these transfer learning algorithms from PLMs, additive learning that incorporates a task-specific adapter on top of the fixed PLM has been popularly used to alleviate the severe overfitting problem in the few-shot setting. However, this added task-specific adapter is generally trained by maximum likelihood estimation that can easily suffer from the so-called exposure bias problem, especially in sequential text generation. Therefore, in this work, we develop a novel additive learning algorithm based on reinforcement learning (RL) for few-shot natural language generation (NLG) tasks. In particular, we propose to use a selective token generation between the transformer-based PLM and the task-specific adapter during both training and inference. This output token selection between the two generators allows the adapter to take into account only on the task-relevant parts in sequence generation, and therefore makes it more robust to overfitting as well as more stable in RL training. In addition, in order to obtain the complementary adapter from the PLM for each few-shot task, we exploit a separate selecting module that is also simultaneously trained using RL. Experimental results on various few-shot NLG tasks including data-to-text generation and text summarization demonstrate that the proposed selective token generation significantly outperforms the previous additive learning algorithms based on the PLMs. | Daejin Jo, Taehwan Kwon, Sungwoong Kim, Eun-Sol Kim | null | null | 2,022 | iclr |
From Intervention to Domain Transportation: A Novel Perspective to Optimize Recommendation | null | The interventional nature of recommendation has attracted increasing attention in recent years. It particularly motivates researchers to formulate learning and evaluating recommendation as causal inference and data missing-not-at-random problems. However, few take seriously the consequence of violating the critical assumption of overlapping, which we prove can significantly threaten the validity and interpretation of the outcome. We find a critical piece missing in the current understanding of information retrieval (IR) systems: as interventions, recommendation not only affects the already observed data, but it also interferes with the target domain (distribution) of interest. We then rephrase optimizing recommendation as finding an intervention that best transports the patterns it learns from the observed domain to its intervention domain. Towards this end, we use domain transportation to characterize the learning-intervention mechanism of recommendation. We design a principled transportation-constraint risk minimization objective and convert it to a two-player minimax game.
We prove the consistency, generalization, and excessive risk bounds for the proposed objective, and elaborate how they compare to the current results. Finally, we carry out extensive real-data and semi-synthetic experiments to demonstrate the advantage of our approach, and launch online testing with a real-world IR system. | Da Xu, Yuting Ye, Chuanwei Ruan, Evren Korpeoglu, Sushant Kumar, Kannan Achan | null | null | 2,022 | iclr |
AdaMatch: A Unified Approach to Semi-Supervised Learning and Domain Adaptation | null | We extend semi-supervised learning to the problem of domain adaptation to learn significantly higher-accuracy models that train on one data distribution and test on a different one. With the goal of generality, we introduce AdaMatch, a unified solution for unsupervised domain adaptation (UDA), semi-supervised learning (SSL), and semi-supervised domain adaptation (SSDA). In an extensive experimental study, we compare its behavior with respective state-of-the-art techniques from SSL, SSDA, and UDA and find that AdaMatch either matches or significantly exceeds the state-of-the-art in each case using the same hyper-parameters regardless of the dataset or task. For example, AdaMatch nearly doubles the accuracy compared to that of the prior state-of-the-art on the UDA task for DomainNet and even exceeds the accuracy of the prior state-of-the-art obtained with pre-training by 6.4% when AdaMatch is trained completely from scratch. Furthermore, by providing AdaMatch with just one labeled example per class from the target domain (i.e., the SSDA setting), we increase the target accuracy by an additional 6.1%, and with 5 labeled examples, by 13.6%. | David Berthelot, Rebecca Roelofs, Kihyuk Sohn, Nicholas Carlini, Alexey Kurakin | null | null | 2,022 | iclr |
Towards Training Billion Parameter Graph Neural Networks for Atomic Simulations | null | Recent progress in Graph Neural Networks (GNNs) for modeling atomic simulations has the potential to revolutionize catalyst discovery, which is a key step in making progress towards the energy breakthroughs needed to combat climate change. However, the GNNs that have proven most effective for this task are memory intensive as they model higher-order interactions in the graphs such as those between triplets or quadruplets of atoms, making it challenging to scale these models. In this paper, we introduce Graph Parallelism, a method to distribute input graphs across multiple GPUs, enabling us to train very large GNNs with hundreds of millions or billions of parameters. We empirically evaluate our method by scaling up the recently proposed DimeNet++ and GemNet models by over an order of magnitude in the number of parameters. On the large-scale Open Catalyst 2020 (OC20) dataset, these graph-parallelized models lead to relative improvements of 1) 15% on the force MAE metric on the S2EF task and 2) 21% on the AFbT metric on the IS2RS task, establishing new state-of-the-art results. | Anuroop Sriram, Abhishek Das, Brandon M Wood, Siddharth Goyal, C. Lawrence Zitnick | null | null | 2,022 | iclr |
Pyraformer: Low-Complexity Pyramidal Attention for Long-Range Time Series Modeling and Forecasting | null | Accurate prediction of the future given the past based on time series data is of paramount importance, since it opens the door for decision making and risk management ahead of time. In practice, the challenge is to build a flexible but parsimonious model that can capture a wide range of temporal dependencies. In this paper, we propose Pyraformer by exploring the multiresolution representation of the time series. Specifically, we introduce the pyramidal attention module (PAM) in which the inter-scale tree structure summarizes features at different resolutions and the intra-scale neighboring connections model the temporal dependencies of different ranges. Under mild conditions, the maximum length of the signal traversing path in Pyraformer is a constant (i.e., $\mathcal O(1)$) with regard to the sequence length $L$, while its time and space complexity scale linearly with $L$. Extensive numerical results show that Pyraformer typically achieves the highest prediction accuracy in both single-step and long-range forecasting tasks with the least amount of time and memory consumption, especially when the sequence is long. | Shizhan Liu, Hang Yu, Cong Liao, Jianguo Li, Weiyao Lin, Alex X. Liu, Schahram Dustdar | null | null | 2,022 | iclr |
Embedding models through the lens of Stable Coloring | null | Embedding-based approaches find the semantic meaning of tokens in structured data such as natural language, graphs, and even images. To a great degree, these approaches have developed independently in different domains. However, we find a common principle underlying these formulations, and it is rooted in solutions to the stable coloring problem in graphs (Weisfeiler-Lehman isomorphism test). For instance, we find links between stable coloring, distribution hypothesis in natural language processing, and non-local-means denoising algorithm in image signal processing. We even find that stable coloring has strong connections to a broad class of unsupervised embedding models which is surprising at first since stable coloring is generally applied for combinatorial problems. To establish this connection concretely we define a mathematical framework that defines continuous stable coloring on graphs and develops optimization problems to search for them. Grounded on this framework, we show that many algorithms ranging across different domains are, in fact, searching for continuous stable coloring solutions of an underlying graph corresponding to the domain. We show that popular and widely used embedding models such as Word2Vec, AWE, BERT, Node2Vec, and Vis-Transformer can be understood as instantiations of our general algorithm that solves the problem of continuous stable coloring. These instantiations offer useful insights into the workings of state-of-the-art models like BERT stimulating new research directions. | Aditya Desai, Shashank Sonkar, Anshumali Shrivastava, Richard Baraniuk | null | null | 2,022 | iclr |
Continual Learning with Recursive Gradient Optimization | null | Learning multiple tasks sequentially without forgetting previous knowledge, called Continual Learning(CL), remains a long-standing challenge for neural networks. Most existing methods rely on additional network capacity or data replay. In contrast, we introduce a novel approach which we refer to as Recursive Gradient Optimization(RGO). RGO is composed of an iteratively updated optimizer that modifies the gradient to minimize forgetting without data replay and a virtual Feature Encoding Layer(FEL) that represents different long-term structures with only task descriptors. Experiments demonstrate that RGO has significantly better performance on popular continual classification benchmarks when compared to the baselines and achieves new state-of-the-art performance on 20-split-CIFAR100(82.22%) and 20-split-miniImageNet(72.63%). With higher average accuracy than Single-Task Learning(STL), this method is flexible and reliable to provide continual learning capabilities for learning models that rely on gradient descent. | Hao Liu, Huaping Liu | null | null | 2,022 | iclr |
GrASP: Gradient-Based Affordance Selection for Planning | null | Planning with a learned model is arguably a key component of intelligence. There are several challenges in realizing such a component in large-scale reinforcement learning (RL) problems. One such challenge is dealing effectively with continuous action spaces when using tree-search planning (e.g., it is not feasible to consider every action even at just the root node of the tree). In this paper we present a method for \emph{selecting} affordances useful for planning---for learning which small number of actions/options from a continuous space of actions/options to consider in the tree-expansion process during planning. We consider affordances that are goal-and-state-conditional mappings to actions/options as well as unconditional affordances that simply select actions/options available in all states. Our selection method is gradient based: we compute gradients through the planning procedure to update the parameters of the function that represents affordances. Our empirical work shows that it is feasible to learn to select both primitive-action and option affordances, and that simultaneously learning to select affordances and planning with a learned value-equivalent model can outperform model-free RL. | Vivek Veeriah, Zeyu Zheng, Richard Lewis, Satinder Singh | null | null | 2,022 | iclr |
F8Net: Fixed-Point 8-bit Only Multiplication for Network Quantization | null | Neural network quantization is a promising compression technique to reduce memory footprint and save energy consumption, potentially leading to real-time inference. However, there is a performance gap between quantized and full-precision models. To reduce it, existing quantization approaches require high-precision INT32 or full-precision multiplication during inference for scaling or dequantization. This introduces a noticeable cost in terms of memory, speed, and required energy. To tackle these issues, we present F8Net, a novel quantization framework consisting in only fixed-point 8-bit multiplication. To derive our method, we first discuss the advantages of fixed-point multiplication with different formats of fixed-point numbers and study the statistical behavior of the associated fixed-point numbers. Second, based on the statistical and algorithmic analysis, we apply different fixed-point formats for weights and activations of different layers. We introduce a novel algorithm to automatically determine the right format for each layer during training. Third, we analyze a previous quantization algorithm—parameterized clipping activation (PACT)—and reformulate it using fixed-point arithmetic. Finally, we unify the recently proposed method for quantization fine-tuning and our fixed-point approach to show the potential of our method. We verify F8Net on ImageNet for MobileNet V1/V2 and ResNet18/50. Our approach achieves comparable and better performance, when compared not only to existing quantization techniques with INT32 multiplication or floating point arithmetic, but also to the full-precision counterparts, achieving state-of-the-art performance. | Qing Jin, Jian Ren, Richard Zhuang, Sumant Hanumante, Zhengang Li, Zhiyu Chen, Yanzhi Wang, Kaiyuan Yang, Sergey Tulyakov | null | null | 2,022 | iclr |
Learning meta-features for AutoML | null | This paper tackles the AutoML problem, aimed to automatically select an ML algorithm and its hyper-parameter configuration most appropriate to the dataset at hand. The proposed approach, MetaBu, learns new meta-features via an Optimal Transport procedure, aligning the manually designed \mf s with the space of distributions on the hyper-parameter configurations. MetaBu meta-features, learned once and for all, induce a topology on the set of datasets that is exploited to define a distribution of promising hyper-parameter configurations amenable to AutoML. Experiments on the OpenML CC-18 benchmark demonstrate that using MetaBu meta-features boosts the performance of state of the art AutoML systems, AutoSklearn (Feurer et al. 2015) and Probabilistic Matrix Factorization (Fusi et al. 2018). Furthermore, the inspection of MetaBu meta-features gives some hints into when an ML algorithm does well. Finally, the topology based on MetaBu meta-features enables to estimate the intrinsic dimensionality of the OpenML benchmark w.r.t. a given ML algorithm or pipeline. The source code is available at https://github.com/luxusg1/metabu. | Herilalaina Rakotoarison, Louisot Milijaona, Andry RASOANAIVO, Michele Sebag, Marc Schoenauer | null | null | 2,022 | iclr |
L0-Sparse Canonical Correlation Analysis | null | Canonical Correlation Analysis (CCA) models are powerful for studying the associations between two sets of variables. The canonically correlated representations, termed \textit{canonical variates} are widely used in unsupervised learning to analyze unlabeled multi-modal registered datasets. Despite their success, CCA models may break (or overfit) if the number of variables in either of the modalities exceeds the number of samples. Moreover, often a significant fraction of the variables measures modality-specific information, and thus removing them is beneficial for identifying the \textit{canonically correlated variates}. Here, we propose $\ell_0$-CCA, a method for learning correlated representations based on sparse subsets of variables from two observed modalities.
Sparsity is obtained by multiplying the input variables by stochastic gates, whose parameters are learned together with the CCA weights via an $\ell_0$-regularized correlation loss.
We further propose $\ell_0$-Deep CCA for solving the problem of non-linear sparse CCA by modeling the correlated representations using deep nets. We demonstrate the efficacy of the method using several synthetic and real examples. Most notably, by gating nuisance input variables, our approach improves the extracted representations compared to other linear, non-linear and sparse CCA-based models. | Ofir Lindenbaum, Moshe Salhov, Amir Averbuch, Yuval Kluger | null | null | 2,022 | iclr |
Latent Variable Sequential Set Transformers for Joint Multi-Agent Motion Prediction | null | Robust multi-agent trajectory prediction is essential for the safe control of robotic systems. A major challenge is to efficiently learn a representation that approximates the true joint distribution of contextual, social, and temporal information to enable planning. We propose Latent Variable Sequential Set Transformers which are encoder-decoder architectures that generate scene-consistent multi-agent trajectories. We refer to these architectures as “AutoBots”. The encoder is a stack of interleaved temporal and social multi-head self-attention (MHSA) modules which alternately perform equivariant processing across the temporal and social dimensions. The decoder employs learnable seed parameters in combination with temporal and social MHSA modules allowing it to perform inference over the
entire future scene in a single forward pass efficiently. AutoBots can produce either the trajectory of one ego-agent or a distribution over the future trajectories for all agents in the scene. For the single-agent prediction case, our model achieves top results on the global nuScenes vehicle motion prediction leaderboard, and produces strong results on the Argoverse vehicle prediction challenge. In the multi-agent setting, we evaluate on the synthetic partition of TrajNet++ dataset to showcase the model’s socially-consistent predictions. We also demonstrate our model on general sequences of sets and provide illustrative experiments modelling the sequential structure of the multiple strokes that make up symbols in the Omniglot data. A distinguishing feature of AutoBots is that all models are trainable on a
single desktop GPU (1080 Ti) in under 48h. | Roger Girgis, Florian Golemo, Felipe Codevilla, Martin Weiss, Jim Aldon D'Souza, Samira Ebrahimi Kahou, Felix Heide, Christopher Pal | null | null | 2,022 | iclr |
Tesseract: Gradient Flip Score to Secure Federated Learning against Model Poisoning Attacks | null | Federated learning—multi-party, distributed learning in a decentralized environment—is vulnerable to model poisoning attacks, even more so than centralized learning approaches. This is because malicious clients can collude and send in carefully tailored model updates to make the global model inaccurate. This motivated the development of Byzantine-resilient federated learning algorithms, such as Krum, Trimmed mean, and FoolsGold. However, a recently developed targeted model poisoning attack showed that all prior defenses can be bypassed. The attack uses the intuition that simply by changing the sign of the gradient updates that the optimizer is computing, for a set of malicious clients, a model can be pushed away from the optima to increase the test error rate. In this work, we develop tesseract—a defense against this directed deviation attack, a state-of-the-art model poisoning attack. TESSERACT is based on a simple intuition that in a federated learning setting, certain patterns of gradient flips are indicative of an attack. This intuition is remarkably stable across different learning algorithms, models, and datasets. TESSERACT assigns reputation scores to the participating clients based on their behavior during the training phase and then takes a weighted contribution of the clients. We show that TESSERACT provides robustness against even an adaptive white-box version of the attack. | Atul Sharma, Wei Chen, Joshua Christian Zhao, Qiang Qiu, Somali Chaterji, Saurabh Bagchi | null | null | 2,022 | iclr |
Non-Transferable Learning: A New Approach for Model Ownership Verification and Applicability Authorization | null | As Artificial Intelligence as a Service gains popularity, protecting well-trained models as intellectual property is becoming increasingly important. There are two common types of protection methods: ownership verification and usage authorization. In this paper, we propose Non-Transferable Learning (NTL), a novel approach that captures the exclusive data representation in the learned model and restricts the model generalization ability to certain domains. This approach provides effective solutions to both model verification and authorization. Specifically: 1) For ownership verification, watermarking techniques are commonly used but are often vulnerable to sophisticated watermark removal methods. By comparison, our NTL-based ownership verification provides robust resistance to state-of-the-art watermark removal methods, as shown in extensive experiments with 6 removal approaches over the digits, CIFAR10 & STL10, and VisDA datasets. 2) For usage authorization, prior solutions focus on authorizing specific users to access the model, but authorized users can still apply the model to any data without restriction. Our NTL-based authorization approach instead provides data-centric protection, which we call applicability authorization, by significantly degrading the performance of the model on unauthorized data. Its effectiveness is also shown through experiments on aforementioned datasets. | Lixu Wang, Shichao Xu, Ruiqi Xu, Xiao Wang, Qi Zhu | null | null | 2,022 | iclr |
Neural Network Approximation based on Hausdorff distance of Tropical Zonotopes | null | In this work we theoretically contribute to neural network approximation by providing a novel tropical geometrical viewpoint to structured neural network compression. In particular, we show that the approximation error between two neural networks with ReLU activations and one hidden layer depends on the Hausdorff distance of the tropical zonotopes of the networks. This theorem comes as a first step towards a purely geometrical interpretation of neural network approximation. Based on this theoretical contribution, we propose geometrical methods that employ the K-means algorithm to compress the fully connected parts of ReLU activated deep neural networks. We analyze the error bounds of our algorithms theoretically based on our approximation theorem and evaluate them empirically on neural network compression. Our experiments follow a proof-of-concept strategy and indicate that our geometrical tools achieve improved performance over relevant tropical geometry techniques and can be competitive against non-tropical methods. | Panagiotis Misiakos, Georgios Smyrnis, George Retsinas, Petros Maragos | null | null | 2,022 | iclr |
Understanding Latent Correlation-Based Multiview Learning and Self-Supervision: An Identifiability Perspective | null | Multiple views of data, both naturally acquired (e.g., image and audio) and artificially produced (e.g., via adding different noise to data samples), have proven useful in enhancing representation learning. Natural views are often handled by multiview analysis tools, e.g., (deep) canonical correlation analysis [(D)CCA], while the artificial ones are frequently used in self-supervised learning (SSL) paradigms, e.g., BYOL and Barlow Twins. Both types of approaches often involve learning neural feature extractors such that the embeddings of data exhibit high cross-view correlations. Although intuitive, the effectiveness of correlation-based neural embedding is mostly empirically validated.
This work aims to understand latent correlation maximization-based deep multiview learning from a latent component identification viewpoint. An intuitive generative model of multiview data is adopted, where the views are different nonlinear mixtures of shared and private components. Since the shared components are view/distortion-invariant, representing the data using such components is believed to reveal the identity of the samples effectively and robustly. Under this model, latent correlation maximization is shown to guarantee the extraction of the shared components across views (up to certain ambiguities). In addition, it is further shown that the private information in each view can be provably disentangled from the shared using proper regularization design. A finite sample analysis, which has been rare in nonlinear mixture identifiability study, is also presented. The theoretical results and newly designed regularization are tested on a series of tasks. | Qi Lyu, Xiao Fu, Weiran Wang, Songtao Lu | null | null | 2,022 | iclr |
I-PGD-AT: Efficient Adversarial Training via Imitating Iterative PGD Attack | null | Adversarial training has been widely used in various machine learning paradigms to improve the robustness; while it would increase the training cost due to the perturbation optimization process. To improve the efficiency, recent studies leverage Fast Gradient Sign Method with Random Start (FGSM-RS) for adversarial training. However, such methods would lead to relatively low robustness and catastrophic overfitting, which means the robustness against iterative attacks (e.g. Projected Gradient Descent (PGD)) would suddenly drop to 0%. Different approaches have been proposed to address this problem, while later studies show that catastrophic overfitting still remains. In this paper, motivated by the fact that expensive iterative adversarial training methods achieve high robustness without catastrophic overfitting, we aim to ask: Can we perform iterative adversarial training in an efficient way? To this end, we first analyze the difference of perturbation generated by FGSM-RS and PGD and find that PGD tends to craft diverse discrete values instead of $\pm 1$ in FGSM-RS. Based on this observation, we propose an efficient single-step adversarial training method I-PGD-AT by adopting I-PGD attack for training, in which I-PGD imitates PGD virtually. Unlike FGSM that crafts the perturbation directly using the sign of gradient, I-PGD imitates the perturbation of PGD based on the magnitude of gradient. Extensive empirical evaluations on CIFAR-10 and Tiny ImageNet demonstrate that our I-PGD-AT can improve the robustness compared with the baselines and significantly delay catastrophic overfitting. Moreover, we explore and discuss the factors that affect catastrophic overfitting. Finally, to demonstrate the generality of I-PGD-AT, we integrate it into PGD adversarial training and show that it can even further improve the robustness. | Xiaosen Wang, Bhavya Kailkhura, Krishnaram Kenthapadi, Bo Li | null | null | 2,022 | iclr |
ARMCMC: Online Bayesian Density Estimation of Model Parameters | null | Although the Bayesian paradigm provides a rigorous framework to estimate the full probability distribution over unknown parameters, its online implementation can be challenging due to heavy computational costs. This paper proposes Adaptive Recursive Markov Chain Monte Carlo (ARMCMC) which estimates full probability density of model parameters while alleviating shortcomings of conventional online approaches. These shortcomings include: being solely able to account for Gaussian noise, being applicable to systems with linear in the parameters (LIP) constraint, or having requirements on persistence excitation (PE). In ARMCMC, we propose a variable jump distribution, which depends on a temporal forgetting factor. This allows one to adjust the trade-off between exploitation and exploration, depending on whether there is an abrupt change to the parameter being estimated. We prove that ARMCMC requires fewer samples to achieve the same precision and reliability compared to conventional MCMC approaches. We demonstrate our approach on two challenging benchmarks: the estimation of parameters in a soft bending actuator and the Hunt-Crossley dynamic model. Our method shows at-least 70\% improvement in parameter point estimation accuracy and approximately 55\% reduction in tracking error of the value of interest compared to recursive least squares and conventional MCMC. | Pedram Agand, Mo Chen, Hamid Taghirad | null | null | 2,022 | iclr |
Geometric and Physical Quantities improve E(3) Equivariant Message Passing | null | Including covariant information, such as position, force, velocity or spin is important in many tasks in computational physics and chemistry. We introduce Steerable E($3$) Equivariant Graph Neural Networks (SEGNNs) that generalise equivariant graph networks, such that node and edge attributes are not restricted to invariant scalars, but can contain covariant information, such as vectors or tensors. Our model, composed of steerable MLPs, is able to incorporate geometric and physical information in both the message and update functions.
Through the definition of steerable node attributes, the MLPs provide a new class of activation functions for general use with steerable feature fields. We discuss ours and related work through the lens of equivariant non-linear convolutions, which further allows us to pin-point the successful components of SEGNNs: non-linear message aggregation improves upon classic linear (steerable) point convolutions; steerable messages improve upon recent equivariant graph networks that send invariant messages. We demonstrate the effectiveness of our method on several tasks in computational physics and chemistry and provide extensive ablation studies. | Johannes Brandstetter, Rob Hesselink, Elise van der Pol, Erik J Bekkers, Max Welling | null | null | 2,022 | iclr |
Wish you were here: Hindsight Goal Selection for long-horizon dexterous manipulation | null | Complex sequential tasks in continuous-control settings often require agents to successfully traverse a set of ``narrow passages'' in their state space. Solving such tasks with a sparse reward in a sample-efficient manner poses a challenge to modern reinforcement learning (RL) due to the associated long-horizon nature of the problem and the lack of sufficient positive signal during learning.
Various tools have been applied to address this challenge. When available, large sets of demonstrations can guide agent exploration. Hindsight relabelling on the other hand does not require additional sources of information. However, existing strategies explore based on task-agnostic goal distributions, which can render the solution of long-horizon tasks impractical. In this work, we extend hindsight relabelling mechanisms to guide exploration along task-specific distributions implied by a small set of successful demonstrations. We evaluate the approach on four complex, single and dual arm, robotics manipulation tasks against strong suitable baselines. The method requires far fewer demonstrations to solve all tasks and achieves a significantly higher overall performance as task complexity increases. Finally, we investigate the robustness of the proposed solution with respect to the quality of input representations and the number of demonstrations. | Todor Davchev, Oleg Olegovich Sushkov, Jean-Baptiste Regli, Stefan Schaal, Yusuf Aytar, Markus Wulfmeier, Jon Scholz | null | null | 2,022 | iclr |
CrossBeam: Learning to Search in Bottom-Up Program Synthesis | null | Many approaches to program synthesis perform a search within an enormous space of programs to find one that satisfies a given specification. Prior works have used neural models to guide combinatorial search algorithms, but such approaches still explore a huge portion of the search space and quickly become intractable as the size of the desired program increases. To tame the search space blowup, we propose training a neural model to learn a hands-on search policy for bottom-up synthesis, instead of relying on a combinatorial search algorithm. Our approach, called CrossBeam, uses the neural model to choose how to combine previously-explored programs into new programs, taking into account the search history and partial program executions. Motivated by work in structured prediction on learning to search, CrossBeam is trained on-policy using data extracted from its own bottom-up searches on training tasks. We evaluate CrossBeam in two very different domains, string manipulation and logic programming. We observe that CrossBeam learns to search efficiently, exploring much smaller portions of the program space compared to the state-of-the-art.
| Kensen Shi, Hanjun Dai, Kevin Ellis, Charles Sutton | null | null | 2,022 | iclr |
Relational Multi-Task Learning: Modeling Relations between Data and Tasks | null | A key assumption in multi-task learning is that at the inference time the multi-task model only has access to a given data point but not to the data point’s labels from other tasks. This presents an opportunity to extend multi-task learning to utilize data point’s labels from other auxiliary tasks, and this way improves performance on the new task. Here we introduce a novel relational multi-task learning setting where we leverage data point labels from auxiliary tasks to make more accurate predictions on the new task. We develop MetaLink, where our key innovation is to build a knowledge graph that connects data points and tasks and thus allows us to leverage labels from auxiliary tasks. The knowledge graph consists of two types of nodes: (1) data nodes, where node features are data embeddings computed by the neural network, and (2) task nodes, with the last layer’s weights for each task as node features. The edges in this knowledge graph capture data-task relationships, and the edge label captures the label of a data point on a particular task. Under MetaLink, we reformulate the new task as a link label prediction problem between a data node and a task node. The MetaLink framework provides flexibility to model knowledge transfer from auxiliary task labels to the task of interest. We evaluate MetaLink on 6 benchmark datasets in both biochemical and vision domains. Experiments demonstrate that MetaLink can successfully utilize the relations among different tasks, outperforming the state-of-the-art methods under the proposed relational multi-task learning setting, with up to 27% improvement in ROC AUC. | Kaidi Cao, Jiaxuan You, Jure Leskovec | null | null | 2,022 | iclr |
Resonance in Weight Space: Covariate Shift Can Drive Divergence of SGD with Momentum | null | Most convergence guarantees for stochastic gradient descent with momentum (SGDm) rely on iid sampling. Yet, SGDm is often used outside this regime, in settings with temporally correlated input samples such as continual learning and reinforcement learning. Existing work has shown that SGDm with a decaying step-size can converge under Markovian temporal correlation. In this work, we show that SGDm under covariate shift with a fixed step-size can be unstable and diverge. In particular, we show SGDm under covariate shift is a parametric oscillator, and so can suffer from a phenomenon known as resonance. We approximate the learning system as a time varying system of ordinary differential equations, and leverage existing theory to characterize the system's divergence/convergence as resonant/nonresonant modes. The theoretical result is limited to the linear setting with periodic covariate shift, so we empirically supplement this result to show that resonance phenomena persist even under non-periodic covariate shift, nonlinear dynamics with neural networks, and optimizers other than SGDm. | Kirby Banman, Garnet Liam Peet-Pare, Nidhi Hegde, Alona Fyshe, Martha White | null | null | 2,022 | iclr |
Recycling Model Updates in Federated Learning: Are Gradient Subspaces Low-Rank? | null | In this paper, we question the rationale behind propagating large numbers of parameters through a distributed system during federated learning. We start by examining the rank characteristics of the subspace spanned by gradients (i.e., the gradient-space) in centralized model training, and observe that the gradient-space often consists of a few leading principal components accounting for an overwhelming majority (95-99%) of the explained variance. Motivated by this, we propose the "Look-back Gradient Multiplier" (LBGM) algorithm, which utilizes this low-rank property of the gradient-space in federated learning. Operationally, LBGM recycles the gradients between model update rounds to significantly reduce the number of parameters to be propagated through the system. We analytically characterize the convergence behavior of LBGM, revealing the nature of the trade-off between communication savings and model performance. Our subsequent experimental results demonstrate the improvement LBGM obtains on communication overhead compared to federated learning baselines. Additionally, we show that LBGM is a general plug-and-play algorithm that can be used standalone or stacked on top of existing sparsification techniques for distributed model training. | Sheikh Shams Azam, Seyyedali Hosseinalipour, Qiang Qiu, Christopher Brinton | null | null | 2,022 | iclr |
Planning in Stochastic Environments with a Learned Model | null | Model-based reinforcement learning has proven highly successful. However, learning a model in isolation from its use during planning is problematic in complex environments. To date, the most effective techniques have instead combined value-equivalent model learning with powerful tree-search methods. This approach is exemplified by MuZero, which has achieved state-of-the-art performance in a wide range of domains, from board games to visually rich environments, with discrete and continuous action spaces, in online and offline settings. However, previous instantiations of this approach were limited to the use of deterministic models. This limits their performance in environments that are inherently stochastic, partially observed, or so large and complex that they appear stochastic to a finite agent. In this paper we extend this approach to learn and plan with stochastic models. Specifically, we introduce a new algorithm, Stochastic MuZero, that learns a stochastic model incorporating afterstates, and uses this model to perform a stochastic tree search. Stochastic MuZero matched or exceeded the state of the art in a set of canonical single and multi-agent environments, including 2048 and backgammon, while maintaining the same performance as standard MuZero in the game of Go. | Ioannis Antonoglou, Julian Schrittwieser, Sherjil Ozair, Thomas K Hubert, David Silver | null | null | 2,022 | iclr |
On the Relationship between Heterophily and Robustness of Graph Neural Networks | null | Empirical studies on the robustness of graph neural networks (GNNs) have suggested a relation between the vulnerabilities of GNNs to adversarial attacks and the increased presence of heterophily in perturbed graphs (where edges tend to connect nodes with dissimilar features and labels). In this work, we formalize the relation between heterophily and robustness, bridging two topics previously investigated by separate lines of research. We theoretically and empirically show that for graphs exhibiting homophily (low heterophily), impactful structural attacks always lead to increased levels of heterophily, while for graph with heterophily the change in the homophily level depends on the node degrees. By leveraging these insights, we deduce that a design principle identified to significantly improve predictive performance under heterophily—separate aggregators for ego- and neighbor-embeddings—can also inherently offer increased robustness to GNNs. Our extensive empirical analysis shows that GNNs adopting this design alone can achieve significantly improved empirical and certifiable robustness compared to the best-performing unvaccinated model. Furthermore, models with this design can be readily combined with explicit defense mechanisms to yield improved robustness with up to 18.33% increase in performance under attacks compared to the best-performing vaccinated model. | Jiong Zhu, Junchen Jin, Donald Loveland, Michael T Schaub, Danai Koutra | null | null | 2,022 | iclr |
Proper Straight-Through Estimator: Breaking symmetry promotes convergence to true minimum | null | In the quantized network, its gradient shows either vanishing or diverging. The network thus cannot be learned by the standard back-propagation, so that an alternative approach called Straight Through Estimator (STE), which replaces the part of the gradient with a simple differentiable function, is used. While STE is known to work well for learning the quantized network empirically, it has not been established theoretically. A recent study by Yin et. al. (2019) has provided theoretical support for STE. However, its justification is still limited to the model in the one-hidden layer network with the binary activation where Gaussian generates the input data, and the true labels are output from the teacher network with the same binary network architecture. In this paper, we discuss the effectiveness of STEs in more general situations without assuming the shape of the input distribution and the labels. By considering the scale symmetry of the network and specific properties of the STEs, we find that STE with clipped Relu is superior to STEs with identity function and vanilla Relu. The clipped Relu STE, which breaks the scale symmetry, may pick up one of the local minima degenerated in scales, while the identity STE and vanilla Relu STE, which keep the scale symmetry, may not pick it up. To confirm this observation, we further present an analysis of a simple misspecified model as an example. We find that all the stationary points are identical with the vanishing points of the cRelu STE gradient, while some of them are not identical with the vanishing points of the identity and Relu STE. | Shinya Gongyo, Kohta Ishikawa | null | null | 2,022 | iclr |
Adversarial Support Alignment | null | We study the problem of aligning the supports of distributions. Compared to the existing work on distribution alignment, support alignment does not require the densities to be matched. We propose symmetric support difference as a divergence measure to quantify the mismatch between supports. We show that select discriminators (e.g. discriminator trained for Jensen-Shannon divergence) are able to map support differences as support differences in their one-dimensional output space. Following this result, our method aligns supports by minimizing a symmetrized relaxed optimal transport cost in the discriminator 1D space via an adversarial process. Furthermore, we show that our approach can be viewed as a limit of existing notions of alignment by increasing transportation assignment tolerance. We quantitatively evaluate the method across domain adaptation tasks with shifts in label distributions. Our experiments show that the proposed method is more robust against these shifts than other alignment-based baselines. | Shangyuan Tong, Timur Garipov, Yang Zhang, Shiyu Chang, Tommi S. Jaakkola | null | null | 2,022 | iclr |
GSD: Generalized Stochastic Decoding | null | Although substantial progress has been made in various text generation tasks, there remains a vast gap between current generations and human languages. One reason is that virtually all decoding methods currently developed are pragmatic to address the text degeneration problem, which exists in both deterministic and stochastic decoding algorithms. So, why text generated from these algorithms are divergent? What is the critical difference between these algorithms? Moreover, is it possible to design a generalized framework where existing decoding algorithms can be naturally connected, uniformly described, and mutually inspired?
In this paper, we try to explore answers to these intriguing questions. Correctly, we propose a generalized decoding framework that can be used to describe and connect existing popular decoding algorithms. Based on the framework, we propose a novel implementation with a distinctive core from existing decoding algorithms. As far as we know, this is the first work trying to propose a generalized framework to bridge these decoding algorithms using formal theorems and concrete implementations. By setting up different conditions, our framework provides infinite space to develop new decoding algorithms. Experiments show that text produced by our method is closest to the characteristics of human languages. Source code and the generated text can be accessed from https://github.com/ginoailab/gsd.git. | Ning Gong, Nianmin Yao | null | null | 2,022 | iclr |
Asymmetry Learning for Counterfactually-invariant Classification in OOD Tasks | null | Generalizing from observed to new related environments (out-of-distribution) is central to the reliability of classifiers. However, most classifiers fail to predict label $Y$ from input $X$ when the change in environment is due a (stochastic) input transformation $T^\text{te} \circ X'$ not observed in training, as in training we observe $T^\text{tr} \circ X'$, where $X'$ is a hidden variable. This work argues that when the transformations in train $T^\text{tr}$ and test $T^\text{te}$ are (arbitrary) symmetry transformations induced by a collection of known $m$ equivalence relations, the task of finding a robust OOD classifier can be defined as finding the simplest causal model that defines a causal connection between the target labels and the symmetry transformations that are associated with label changes. We then propose a new learning paradigm, asymmetry learning, that identifies which symmetries the classifier must break in order to correctly predict $Y$ in both train and test. Asymmetry learning performs a causal model search that, under certain identifiability conditions, finds classifiers that perform equally well in-distribution and out-of-distribution. Finally, we show how to learn counterfactually-invariant representations with asymmetry learning in two physics tasks. | S Chandra Mouli, Bruno Ribeiro | null | null | 2,022 | iclr |
Variational Inference for Discriminative Learning with Generative Modeling of Feature Incompletion | null | We are concerned with the problem of distributional prediction with incomplete features: The goal is to estimate the distribution of target variables given feature vectors with some of the elements missing. A typical approach to this problem is to perform missing-value imputation and regression, simultaneously or sequentially, which we call the generative approach. Another approach is to perform regression after appropriately encoding missing values into the feature, which we call the discriminative approach. In comparison, the generative approach is more robust to the feature corruption while the discriminative approach is more favorable to maximize the performance of prediction.
In this study, we propose a hybrid method to take the best of both worlds. Our method utilizes the black-box variational inference framework so that it can be applied to a wide variety of modern machine learning models, including the variational autoencoders. We also confirmed the effectiveness of the proposed method empirically.
| Kohei Miyaguchi, Takayuki Katsuki, Akira Koseki, Toshiya Iwamori | null | null | 2,022 | iclr |
Approximation and Learning with Deep Convolutional Models: a Kernel Perspective | null | The empirical success of deep convolutional networks on tasks involving high-dimensional data such as images or audio suggests that they can efficiently approximate certain functions that are well-suited for such tasks. In this paper, we study this through the lens of kernel methods, by considering simple hierarchical kernels with two or three convolution and pooling layers, inspired by convolutional kernel networks. These achieve good empirical performance on standard vision datasets, while providing a precise description of their functional space that yields new insights on their inductive bias. We show that the RKHS consists of additive models of interaction terms between patches, and that its norm encourages spatial similarities between these terms through pooling layers. We then provide generalization bounds which illustrate how pooling and patches yield improved sample complexity guarantees when the target function presents such regularities. | Alberto Bietti | null | null | 2,022 | iclr |
The Hidden Convex Optimization Landscape of Regularized Two-Layer ReLU Networks: an Exact Characterization of Optimal Solutions | null | We prove that finding all globally optimal two-layer ReLU neural networks can be performed by solving a convex optimization program with cone constraints. Our analysis is novel, characterizes all optimal solutions, and does not leverage duality-based analysis which was recently used to lift neural network training into convex spaces. Given the set of solutions of our convex optimization program, we show how to construct exactly the entire set of optimal neural networks. We provide a detailed characterization of this optimal set and its invariant transformations. As additional consequences of our convex perspective, (i) we establish that Clarke stationary points found by stochastic gradient descent correspond to the global optimum of a subsampled convex problem (ii) we provide a polynomial-time algorithm for checking if a neural network is a global minimum of the training loss (iii) we provide an explicit construction of a continuous path between any neural network and the global minimum of its sublevel set and (iv) characterize the minimal size of the hidden layer so that the neural network optimization landscape has no spurious valleys.
Overall, we provide a rich framework for studying the landscape of neural network training loss through convexity. | Yifei Wang, Jonathan Lacotte, Mert Pilanci | null | null | 2,022 | iclr |
Unsupervised Vision-Language Grammar Induction with Shared Structure Modeling | null | We introduce a new task, unsupervised vision-language (VL) grammar induction. Given an image-caption pair, the goal is to extract a shared hierarchical structure for both image and language simultaneously. We argue that such structured output, grounded in both modalities, is a clear step towards the high-level understanding of multimodal information. Besides challenges existing in conventional visually grounded grammar induction tasks, VL grammar induction requires a model to capture contextual semantics and perform a fine-grained alignment. To address these challenges, we propose a novel method, CLIORA, which constructs a shared vision-language constituency tree structure with context-dependent semantics for all possible phrases in different levels of the tree. It computes a matching score between each constituent and image region, trained via contrastive learning. It integrates two levels of fusion, namely at feature-level and at score-level, so as to allow fine-grained alignment. We introduce a new evaluation metric for VL grammar induction, CCRA, and show a 3.3% improvement over a strong baseline on Flickr30k Entities. We also evaluate our model via two derived tasks, i.e., language grammar induction and phrase grounding, and improve over the state-of-the-art for both. | Bo Wan, Wenjuan Han, Zilong Zheng, Tinne Tuytelaars | null | null | 2,022 | iclr |
SOSP: Efficiently Capturing Global Correlations by Second-Order Structured Pruning | null | Pruning neural networks reduces inference time and memory costs. On standard hardware, these benefits will be especially prominent if coarse-grained structures, like feature maps, are pruned. We devise two novel saliency-based methods for second-order structured pruning (SOSP) which include correlations among all structures and layers. Our main method SOSP-H employs an innovative second-order approximation, which enables saliency evaluations by fast Hessian-vector products. SOSP-H thereby scales like a first-order method despite taking into account the full Hessian. We validate SOSP-H by comparing it to our second method SOSP-I that uses a well-established Hessian approximation, and to numerous state-of-the-art methods. While SOSP-H performs on par or better in terms of accuracy, it has clear advantages in terms of scalability and efficiency. This allowed us to scale SOSP-H to large-scale vision tasks, even though it captures correlations across all layers of the network. To underscore the global nature of our pruning methods, we evaluate their performance not only by removing structures from a pretrained network, but also by detecting architectural bottlenecks. We show that our algorithms allow to systematically reveal architectural bottlenecks, which we then remove to further increase the accuracy of the networks. | Manuel Nonnenmacher, Thomas Pfeil, Ingo Steinwart, David Reeb | null | null | 2,022 | iclr |
Multi-Stage Episodic Control for Strategic Exploration in Text Games | null | Text adventure games present unique challenges to reinforcement learning methods due to their combinatorially large action spaces and sparse rewards. The interplay of these two factors is particularly demanding because large action spaces require extensive exploration, while sparse rewards provide limited feedback. This work proposes to tackle the explore-vs-exploit dilemma using a multi-stage approach that explicitly disentangles these two strategies within each episode. Our algorithm, called eXploit-Then-eXplore (XTX), begins each episode using an exploitation policy that imitates a set of promising trajectories from the past, and then switches over to an exploration policy aimed at discovering novel actions that lead to unseen state spaces. This policy decomposition allows us to combine global decisions about which parts of the game space to return to with curiosity-based local exploration in that space, motivated by how a human may approach these games. Our method significantly outperforms prior approaches by 27% and 11% average normalized score over 12 games from the Jericho benchmark (Hausknecht et al., 2020) in both deterministic and stochastic settings, respectively. On the game of Zork1, in particular, XTX obtains a score of 103, more than a 2x improvement over prior methods, and pushes past several known bottlenecks in the game that have plagued previous state-of-the-art methods. | Jens Tuyls, Shunyu Yao, Sham M. Kakade, Karthik R Narasimhan | null | null | 2,022 | iclr |
Word Sense Induction with Knowledge Distillation from BERT | null | Pre-trained contextual language models are ubiquitously employed for language understanding tasks, but are unsuitable for resource-constrained systems. Noncontextual word embeddings are an efficient alternative in these settings. Such methods typically use one vector to encode multiple different meanings of a word, and incur errors due to polysemy. This paper proposes a two-stage method to distill multiple word senses from a pre-trained language model (BERT) by using attention over the senses of a word in a context and transferring this sense information to fit multi-sense embeddings in a skip-gram-like framework. We demonstrate an effective approach to training the sense disambiguation mechanism in our model with a distribution over word senses extracted from the output layer embeddings of BERT. Experiments on the contextual word similarity and sense induction tasks show that this method is superior to or competitive with state-of-the-art multi-sense embeddings on multiple benchmark data sets, and experiments with an embedding-based topic model (ETM) demonstrates the benefits of using this multi-sense embedding in a downstream application.
| Anik Saha, Alex Gittens, Bulent Yener | null | null | 2,022 | iclr |
A Multi-Task Learning Algorithm for Non-personalized Recommendations | null | In this paper, we introduce a multi-task learning (MTL) algorithm for recommending non-personalized videos to watch next on industrial video sharing platforms. Personalized recommendations have been studied for decades, while researches on non-personalized solutions are very rare to be seen, which still remain a huge portion in industry. As an indispensable part in recommender system, non-personalized video recommender system also faces several real-world challenges, including maintaining high relevance between source item and target items, as well as achieving multiple competing ranking objectives. To solve these, we largely extended model-based collaborative filtering algorithm by adding related candidate generation stage, Two-tower DNN structure and a multi-task learning mechanism. Compared with typical baseline solutions, our proposed algorithm can capture both linear and non-linear relationships from user-item interactions, and live experiments demonstrate that it can significantly advance the state of the art on recommendation quality. | Jiawei Zhang | null | null | 2,022 | iclr |
Comparing Distributions by Measuring Differences that Affect Decision Making | null | Measuring the discrepancy between two probability distributions is a fundamental problem in machine learning and statistics. We propose a new class of discrepancies based on the optimal loss for a decision task -- two distributions are different if the optimal decision loss is higher on their mixture than on each individual distribution. By suitably choosing the decision task, this generalizes the Jensen-Shannon divergence and the maximum mean discrepancy family. We apply our approach to two-sample tests, and on various benchmarks, we achieve superior test power compared to competing methods. In addition, a modeler can directly specify their preferences when comparing distributions through the decision loss. We apply this property to understanding the effects of climate change on different social and economic activities, evaluating sample quality, and selecting features targeting different decision tasks. | Shengjia Zhao, Abhishek Sinha, Yutong He, Aidan Perreault, Jiaming Song, Stefano Ermon | null | null | 2,022 | iclr |
A New Perspective on "How Graph Neural Networks Go Beyond Weisfeiler-Lehman?" | null | We propose a new perspective on designing powerful Graph Neural Networks (GNNs). In a nutshell, this enables a general solution to inject structural properties of graphs into a message-passing aggregation scheme of GNNs. As a theoretical basis, we develop a new hierarchy of local isomorphism on neighborhood subgraphs. Then, we theoretically characterize how message-passing GNNs can be designed to be more expressive than the Weisfeiler Lehman test. To elaborate this characterization, we propose a novel neural model, called GraphSNN, and prove that this model is strictly more expressive than the Weisfeiler Lehman test in distinguishing graph structures. We empirically verify the strength of our model on different graph learning tasks. It is shown that our model consistently improves the state-of-the-art methods on the benchmark tasks without sacrificing computational simplicity and efficiency. | Asiri Wijesinghe, Qing Wang | null | null | 2,022 | iclr |
Additive Poisson Process: Learning Intensity of Higher-Order Interaction in Poisson Processes | null | We present the Additive Poisson Process (APP), a novel framework that can model the higher-order interaction effects of the intensity functions in Poisson processes using projections into lower-dimensional space. Our model combines the techniques in information geometry to model higher-order interactions on a statistical manifold and in generalized additive models to use lower-dimensional projections to overcome the effects from the curse of dimensionality. Our approach solves a convex optimization problem by minimizing the KL divergence from a sample distribution in lower-dimensional projections to the distribution modeled by an intensity function in the Poisson process. Our empirical results show that our model is able to use samples observed in the lower dimensional space to estimate the higher-order intensity function with extremely sparse observations. | Simon Luo, Feng Zhou, lamiae azizi, Mahito Sugiyama | null | null | 2,022 | iclr |
Minibatch vs Local SGD with Shuffling: Tight Convergence Bounds and Beyond | null | In distributed learning, local SGD (also known as federated averaging) and its simple baseline minibatch SGD are widely studied optimization methods. Most existing analyses of these methods assume independent and unbiased gradient estimates obtained via with-replacement sampling. In contrast, we study shuffling-based variants: minibatch and local Random Reshuffling, which draw stochastic gradients without replacement and are thus closer to practice. For smooth functions satisfying the Polyak-Łojasiewicz condition, we obtain convergence bounds (in the large epoch regime) which show that these shuffling-based variants converge faster than their with-replacement counterparts. Moreover, we prove matching lower bounds showing that our convergence analysis is tight. Finally, we propose an algorithmic modification called synchronized shuffling that leads to convergence rates faster than our lower bounds in near-homogeneous settings. | Chulhee Yun, Shashank Rajput, Suvrit Sra | null | null | 2,022 | iclr |
Deconstructing the Inductive Biases of Hamiltonian Neural Networks | null | Physics-inspired neural networks (NNs), such as Hamiltonian or Lagrangian NNs, dramatically outperform other learned dynamics models by leveraging strong inductive biases. These models, however, are challenging to apply to many real world systems, such as those that don’t conserve energy or contain contacts, a common setting for robotics and reinforcement learning. In this paper, we examine the inductive biases that make physics-inspired models successful in practice. We show that, contrary to conventional wisdom, the improved generalization of HNNs is the result of modeling acceleration directly and avoiding artificial complexity from the coordinate system, rather than symplectic structure or energy conservation. We show that by relaxing the inductive biases of these models, we can match or exceed performance on energy-conserving systems while dramatically improving performance on practical, non-conservative systems. We extend this approach to constructing transition models for common Mujoco environments, showing that our model can appropriately balance inductive biases with the flexibility required for model-based control. | Nate Gruver, Marc Anton Finzi, Samuel Don Stanton, Andrew Gordon Wilson | null | null | 2,022 | iclr |
HFSP: A Hardware-friendly Soft Pruning Framework for Vision Transformers | null | Recently, Vision Transformer (ViT) has continuously established new milestones in the computer vision field, while the high computation and memory cost makes its propagation in industrial production difficult. Pruning, a traditional model compression paradigm for hardware efficiency, has been widely applied in various DNN structures. Nevertheless, it stays ambiguous on how to perform exclusive pruning on the ViT structure. Considering three key points: the structural characteristics, the internal data pattern of ViT, and the related edge device deployment, we leverage the input token sparsity and propose a hardware-friendly soft pruning framework (HFSP), which can be set up on vanilla Transformers of both flatten and CNN-type structures, such as Pooling-based ViT (PiT). More concretely, we design a dynamic attention-based multi-head token selector, which is a lightweight module for adaptive instance-wise token selection. We further introduce a soft pruning technique to package the pruned tokens, which integrate the less informative tokens generated by the selector module into a package token, and participates in subsequent calculations rather than being discarded completely. From a hardware standpoint, our framework is bound to the tradeoff between accuracy and specific hardware constraints through our proposed hardware-oriented progressive training, and all the operators embedded in the framework have been well-supported. Experimental results demonstrate that the proposed framework significantly reduces the computational costs of ViTs while maintaining comparable performance on image classification. For example, our method reduces the FLOPs of DeiT-S by over 42.6% while only sacrificing 0.46% top-1 accuracy. Moreover, our framework can guarantee the identified model to meet resource specifications of mobile devices and FPGA, and even achieve the real-time execution of DeiT-T on mobile platforms. Code will be publicly released. | Zhenglun Kong, Peiyan Dong, Xiaolong Ma, Xin Meng, Mengshu Sun, Wei Niu, Bin Ren, Minghai Qin, Hao Tang, Yanzhi Wang | null | null | 2,022 | iclr |
Transform2Act: Learning a Transform-and-Control Policy for Efficient Agent Design | null | An agent's functionality is largely determined by its design, i.e., skeletal structure and joint attributes (e.g., length, size, strength). However, finding the optimal agent design for a given function is extremely challenging since the problem is inherently combinatorial and the design space is prohibitively large. Additionally, it can be costly to evaluate each candidate design which requires solving for its optimal controller. To tackle these problems, our key idea is to incorporate the design procedure of an agent into its decision-making process. Specifically, we learn a conditional policy that, in an episode, first applies a sequence of transform actions to modify an agent's skeletal structure and joint attributes, and then applies control actions under the new design. To handle a variable number of joints across designs, we use a graph-based policy where each graph node represents a joint and uses message passing with its neighbors to output joint-specific actions. Using policy gradient methods, our approach enables joint optimization of agent design and control as well as experience sharing across different designs, which improves sample efficiency substantially. Experiments show that our approach, Transform2Act, outperforms prior methods significantly in terms of convergence speed and final performance. Notably, Transform2Act can automatically discover plausible designs similar to giraffes, squids, and spiders. Code and videos are available at https://sites.google.com/view/transform2act. | Ye Yuan, Yuda Song, Zhengyi Luo, Wen Sun, Kris M. Kitani | null | null | 2,022 | iclr |
Sample and Computation Redistribution for Efficient Face Detection | null | Although tremendous strides have been made in uncontrolled face detection, accurate face detection with a low computation cost remains an open challenge. In this paper, we point out that computation distribution and scale augmentation are the keys to detecting small faces from low-resolution images. Motivated by these observations, we introduce two simple but effective methods: (1) Computation Redistribution (CR), which reallocates the computation between the backbone, neck and head of the model; and (2) Sample Redistribution (SR), which augments training samples for the most needed stages. The proposed Sample and Computation Redistribution for Face Detection (SCRFD) is implemented by a random search in a meticulously designed search space. Extensive experiments conducted on WIDER FACE demonstrate the state-of-the-art accuracy-efficiency trade-off for the proposed SCRFD family across a wide range of compute regimes. In particular, SCRFD-34GF outperforms the best competitor, TinaFace, by $4.78\%$ (AP at hard set) while being more than 3$\times$ faster on GPUs with VGA-resolution images. Code is available at: https://github.com/deepinsight/insightface/tree/master/detection/scrfd. | Jia Guo, Jiankang Deng, Alexandros Lattas, Stefanos Zafeiriou | null | null | 2,022 | iclr |
Optimal Transport for Causal Discovery | null | To determine causal relationships between two variables, approaches based on Functional Causal Models (FCMs) have been proposed by properly restricting model classes; however, the performance is sensitive to the model assumptions, which makes it difficult to use. In this paper, we provide a novel dynamical-system view of FCMs and propose a new framework for identifying causal direction in the bivariate case. We first show the connection between FCMs and optimal transport, and then study optimal transport under the constraints of FCMs. Furthermore, by exploiting the dynamical interpretation of optimal transport under the FCM constraints, we determine the corresponding underlying dynamical process of the static cause-effect pair data. It provides a new dimension for describing static causal discovery tasks while enjoying more freedom for modeling the quantitative causal influences. In particular, we show that Additive Noise Models (ANMs) correspond to volume-preserving pressureless flows. Consequently, based on their velocity field divergence, we introduce a criterion for determining causal direction. With this criterion, we propose a novel optimal transport-based algorithm for ANMs which is robust to the choice of models and extend it to post-nonlinear models. Our method demonstrated state-of-the-art results on both synthetic and causal discovery benchmark datasets. | Ruibo Tu, Kun Zhang, Hedvig Kjellstrom, Cheng Zhang | null | null | 2,022 | iclr |
Reinforcement Learning under a Multi-agent Predictive State Representation Model: Method and Theory | null | We study reinforcement learning for partially observable multi-agent systems where each agent only has access to its own observation and reward and aims to maximize its cumulative rewards. To handle partial observations, we propose graph-assisted predictive state representations (GAPSR), a scalable multi-agent representation learning framework that leverages the agent connectivity graphs to aggregate local representations computed by each agent. In addition, our representations are readily able to incorporate dynamic interaction graphs and kernel space embeddings of the predictive states, and thus have strong flexibility and representation power.
Based on GAPSR, we propose an end-to-end MARL algorithm that simultaneously infers the predictive representations and uses the representations as the input of a policy optimization algorithm. Empirically, we demonstrate the efficacy of the proposed algorithm provided on both a MAMuJoCo robotic learning experiment and a multi-agent particle learning environment. | Zhi Zhang, Zhuoran Yang, Han Liu, Pratap Tokekar, Furong Huang | null | null | 2,022 | iclr |
Recognizing and overcoming the greedy nature of learning in multi-modal deep neural networks | null | We hypothesize that due to the greedy nature of learning in multi-modal deep neural networks (DNNs), these models tend to rely on just one modality while under-utilizing the other modalities. We observe empirically that such behavior hurts its overall generalization. We validate our hypothesis by estimating the gain on the accuracy when the model has access to an additional modality. We refer to this gain as the conditional utilization rate of the modality. In the experiments, we consistently observe an imbalance in conditional utilization rate between modalities, across multiple tasks and architectures. Since conditional utilization rate cannot be computed efficiently during training, we introduce an efficient proxy based on the pace at which a DNN learns from each modality, which we refer to as conditional learning speed. We thus propose a training algorithm, balanced multi-modal learning, and demonstrate that it indeed addresses the issue of greedy learning. The proposed algorithm is found to improve the model’s generalization on three datasets: Colored MNIST (Kim et al., 2019), Princeton ModelNet40 (Wu et al., 2015), and NVIDIA Dynamic Hand Gesture Dataset (Molchanov et al., 2016). | Nan Wu, Stanislaw Kamil Jastrzebski, Kyunghyun Cho, Krzysztof J. Geras | null | null | 2,022 | iclr |
Understanding and Leveraging Overparameterization in Recursive Value Estimation | null | The theory of function approximation in reinforcement learning (RL) typically considers low capacity representations that incur a tradeoff between approximation error, stability and generalization. Current deep architectures, however, operate in an overparameterized regime where approximation error is not necessarily a bottleneck. To better understand the utility of deep models in RL we present an analysis of recursive value estimation using \emph{overparameterized} linear representations that provides useful, transferable findings. First, we show that classical updates such as temporal difference (TD) learning or fitted-value-iteration (FVI) converge to \emph{different} fixed points than residual minimization (RM) in the overparameterized linear case. We then develop a unified interpretation of overparameterized linear value estimation as minimizing the Euclidean norm of the weights subject to alternative constraints. A practical consequence is that RM can be modified by a simple alteration of the backup targets to obtain the same fixed points as FVI and TD (when they converge), while universally ensuring stability. Further, we provide an analysis of the generalization error of these methods, demonstrating per iterate bounds on the value prediction error of FVI, and fixed point bounds for TD and RM.
Given this understanding, we then develop new algorithmic tools for improving recursive value estimation with deep models.
In particular, we extract two regularizers that penalize out-of-span top-layer weights and co-linearity in top-layer features respectively. Empirically we find that these regularizers dramatically improve the stability of TD and FVI, while allowing RM to match and even sometimes surpass their generalization performance with assured stability. | Chenjun Xiao, Bo Dai, Jincheng Mei, Oscar A Ramirez, Ramki Gummadi, Chris Harris, Dale Schuurmans | null | null | 2,022 | iclr |
Language modeling via stochastic processes | null | Modern language models can generate high-quality short texts. However, they often meander or are incoherent when generating longer texts. These issues arise from the next-token-only language modeling objective. To address these issues, we introduce Time Control (TC), a language model that implicitly plans via a latent stochastic process. TC does this by learning a representation which maps the dynamics of how text changes in a document to the dynamics of a stochastic process of interest. Using this representation, the language model can generate text by first implicitly generating a document plan via a stochastic process, and then generating text that is consistent with this latent plan. Compared to domain-specific methods and fine-tuning GPT2 across a variety of text domains, TC improves performance on text infilling and discourse coherence. On long text generation settings, TC preserves the text structure both in terms of ordering (up to +40% better) and text length consistency (up to +17% better). Human evaluators also prefer TC's output 28.6% more than the baselines. | Rose E Wang, Esin Durmus, Noah Goodman, Tatsunori Hashimoto | null | null | 2,022 | iclr |
Sampling Before Training: Rethinking the Effect of Edges in the Process of Training Graph Neural Networks | null | Graph neural networks (GNN) demonstrate excellent performance on many graph-based tasks; however, they also impose a heavy computational burden when trained on a large-scale graph. Although various sampling methods have been proposed to speed up training GNN by shrinking the scale of the graph during training, they become unavailable if we need to perform sampling before training. In this paper, we quantify the importance of every edge for training in the graph with the extra information they convey in addition to the node features, as inspired by a manifold learning algorithm called diffusion map. Based on this calculation, we propose Graph Diffusion Sampling (GDS), a simple but effective sampling method for shrinking the size of the edge set before training. GDS prefers to sample edges with high importance, and edges dropped by GDS will never be used in the training procedure. We empirically show that GDS preserves the edges crucial for training in a variety of models (GCN, GraphSAGE, GAT, and JKNet). Compared to training on the full graph, GDS can guarantee the performance of the model while only samples a small fraction of the edges.
| Hengyuan Ma, Qi Yang, Bowen Sun, Long Shun, Junkui Li, Jianfeng Feng | null | null | 2,022 | iclr |
Does your graph need a confidence boost? Convergent boosted smoothing on graphs with tabular node features | null | Many practical modeling tasks require making predictions using tabular data composed of heterogeneous feature types (e.g., text-based, categorical, continuous, etc.). In this setting boosted decision trees and related ensembling techniques generally dominate real-world applications involving iid training/test sets. However, when there are relations between samples and the iid assumption is no longer reasonable, it remains unclear how to incorporate these dependencies within existing boosting pipelines. To this end, we propose a generalized framework for combining boosted trees and more general model ensembling techniques, with graph propagation layers that share node/sample information across edges connecting related samples. And unlike previous efforts to integrate graph-based models with boosting, our approach is anchored to a principled meta loss function such that provable convergence can be guaranteed under relatively mild assumptions. Across a variety of benchmarks involving non-iid graph data with tabular node features, our framework achieves comparable or superior performance. | Jiuhai Chen, Jonas Mueller, Vassilis N. Ioannidis, Soji Adeshina, Yangkun Wang, Tom Goldstein, David Wipf | null | null | 2,022 | iclr |
Message Passing Neural PDE Solvers | null | The numerical solution of partial differential equations (PDEs) is difficult, having led to a century of research so far. Recently, there have been pushes to build neural--numerical hybrid solvers, which piggy-backs the modern trend towards fully end-to-end learned systems. Most works so far can only generalize over a subset of properties to which a generic solver would be faced, including: resolution, topology, geometry, boundary conditions, domain discretization regularity, dimensionality, etc. In this work, we build a solver, satisfying these properties, where all the components are based on neural message passing, replacing all heuristically designed components in the computation graph with backprop-optimized neural function approximators. We show that neural message passing solvers representationally contain some classical methods, such as finite differences, finite volumes, and WENO schemes. In order to encourage stability in training autoregressive models, we put forward a method that is based on the principle of zero-stability, posing stability as a domain adaptation problem. We validate our method on various fluid-like flow problems, demonstrating fast, stable, and accurate performance across different domain topologies, discretization, etc. in 1D and 2D. Our model outperforms state-of-the-art numerical solvers in the low resolution regime in terms of speed, and accuracy. | Johannes Brandstetter, Daniel E. Worrall, Max Welling | null | null | 2,022 | iclr |
On strong convergence of the two-tower model for recommender system | null | Recommender system is capable of predicting preferred items for a user by integrating information from similar users or items. A popular model in recommender system is the so-called two-tower model, which employs two deep neural networks to embed users and items into a low-dimensional space, and predicts ratings via the geometrical relationship of the embeddings of user and item in the embedded space. Even though it is popularly used for recommendations, its theoretical properties remain largely unknown. In this paper, we establish some asymptotic results of the two-tower model in terms of its strong convergence to the optimal recommender system, showing that it achieves a fast convergence rate depending on the intrinsic dimensions of inputs features. To the best of our knowledge, this is among the first attempts to establish the statistical guarantee of the two-tower model. Through numerical experiments, we also demonstrate that the two-tower model is capable of capturing the effects of users' and items' features on ratings, leading to higher prediction accuracy over its competitors in both simulated examples and a real application data set. | SHIRONG XU, Junhui Wang | null | null | 2,022 | iclr |
Improving greedy core-set configurations for active learning with uncertainty-scaled distances | null | We scale perceived distances of the core-set algorithm by a factor of uncertainty and search for low-confidence configurations, finding significant improvements in sample efficiency across CIFAR10/100 and SVHN image classification, especially in larger acquisition sizes. We show the necessity of our modifications and explain how the improvement is due to a probabilistic quadratic speed-up in the convergence of core-set loss, under assumptions about the relationship of model uncertainty and misclassification. | Yuchen Li, Frank Rudzicz | null | null | 2,022 | iclr |
Neural Contextual Bandits with Deep Representation and Shallow Exploration | null | We study neural contextual bandits, a general class of contextual bandits, where each context-action pair is associated with a raw feature vector, but the specific reward generating function is unknown. We propose a novel learning algorithm that transforms the raw feature vector using the last hidden layer of a deep ReLU neural network (deep representation learning), and uses an upper confidence bound (UCB) approach to explore in the last linear layer (shallow exploration). We prove that under standard assumptions, our proposed algorithm achieves $\tilde{O}(\sqrt{T})$ finite-time regret, where $T$ is the learning time horizon. Compared with existing neural contextual bandit algorithms, our approach is computationally much more efficient since it only needs to explore in the last layer of the deep neural network. | Pan Xu, Zheng Wen, Handong Zhao, Quanquan Gu | null | null | 2,022 | iclr |
Neural Collapse Under MSE Loss: Proximity to and Dynamics on the Central Path | null | The recently discovered Neural Collapse (NC) phenomenon occurs pervasively in today's deep net training paradigm of driving cross-entropy (CE) loss towards zero. During NC, last-layer features collapse to their class-means, both classifiers and class-means collapse to the same Simplex Equiangular Tight Frame, and classifier behavior collapses to the nearest-class-mean decision rule. Recent works demonstrated that deep nets trained with mean squared error (MSE) loss perform comparably to those trained with CE. As a preliminary, we empirically establish that NC emerges in such MSE-trained deep nets as well through experiments on three canonical networks and five benchmark datasets. We provide, in a Google Colab notebook, PyTorch code for reproducing MSE-NC and CE-NC: https://colab.research.google.com/github/neuralcollapse/neuralcollapse/blob/main/neuralcollapse.ipynb. The analytically-tractable MSE loss offers more mathematical opportunities than the hard-to-analyze CE loss, inspiring us to leverage MSE loss towards the theoretical investigation of NC. We develop three main contributions: (I) We show a new decomposition of the MSE loss into (A) terms directly interpretable through the lens of NC and which assume the last-layer classifier is exactly the least-squares classifier; and (B) a term capturing the deviation from this least-squares classifier. (II) We exhibit experiments on canonical datasets and networks demonstrating that term-(B) is negligible during training. This motivates us to introduce a new theoretical construct: the central path, where the linear classifier stays MSE-optimal for feature activations throughout the dynamics. (III) By studying renormalized gradient flow along the central path, we derive exact dynamics that predict NC. | X.Y. Han, Vardan Papyan, David L. Donoho | null | null | 2,022 | iclr |
VUT: Versatile UI Transformer for Multimodal Multi-Task User Interface Modeling | null | User interface modeling is inherently multimodal, which involves several distinct types of data: images, structures and language. The tasks are also diverse, including object detection, language generation and grounding. In this paper, we present VUT, a Versatile UI Transformer that takes multimodal input and simultaneously accomplishes 5 distinct tasks with the same model. Our model consists of a multimodal Transformer encoder that jointly encodes UI images and structures, and performs UI object detection when the UI structures are absent in the input. Our model also consists of an auto-regressive Transformer model that encodes the language input and decodes output, for both question-answering and command grounding with respect to the UI. Our experiments show that for most of the tasks, when trained jointly for multi-tasks, VUT has achieved accuracy either on par with or exceeding the accuracy when the model is trained for individual tasks separately. | Yang Li, Gang Li, Xin Zhou, Mostafa Dehghani, Alexey A. Gritsenko | null | null | 2,022 | iclr |
Logical Activation Functions: Logit-space equivalents of Boolean Operators | null | Neuronal representations within artificial neural networks are commonly understood as logits, representing the log-odds score of presence (versus absence) of features within the stimulus. Under this interpretation, we can derive the probability $P(x_0 \cap x_1)$ that a pair of independent features are both present in the stimulus from their logits. By converting the resulting probability back into a logit, we obtain a logit-space equivalent of the AND operation. However, since this function involves taking multiple exponents and logarithms, it is not well suited to be directly used within neural networks. We thus constructed an efficient approximation named $\text{AND}_\text{AIL}$ (the AND operator Approximate for Independent Logits) utilizing only comparison and addition operations, which can be deployed as an activation function in neural networks. Like MaxOut, $\text{AND}_\text{AIL}$ is a generalization of ReLU to two-dimensions. Additionally, we constructed efficient approximations of the logit-space equivalents to the OR and XNOR operators. We deployed these new activation functions, both in isolation and in conjunction, and demonstrated their effectiveness on a variety of tasks including image classification, transfer learning, abstract reasoning, and compositional zero-shot learning. | Scott C Lowe, Robert Earle, Jason d'Eon, Thomas Trappenberg, Sageev Oore | null | null | 2,022 | iclr |
CROP: Certifying Robust Policies for Reinforcement Learning through Functional Smoothing | null | As reinforcement learning (RL) has achieved great success and been even adopted in safety-critical domains such as autonomous vehicles, a range of empirical studies have been conducted to improve its robustness against adversarial attacks. However, how to certify its robustness with theoretical guarantees still remains challenging. In this paper, we present the first unified framework CROP (Certifying Robust Policies for RL) to provide robustness certification on both action and reward levels. In particular, we propose two robustness certification criteria: robustness of per-state actions and lower bound of cumulative rewards. We then develop a local smoothing algorithm for policies derived from Q-functions to guarantee the robustness of actions taken along the trajectory; we also develop a global smoothing algorithm for certifying the lower bound of a finite-horizon cumulative reward, as well as a novel local smoothing algorithm to perform adaptive search in order to obtain tighter reward certification. Empirically, we apply CROP to evaluate several existing empirically robust RL algorithms, including adversarial training and different robust regularization, in four environments (two representative Atari games, Highway, and CartPole). Furthermore, by evaluating these algorithms against adversarial attacks, we demonstrate that our certifications are often tight. All experiment results are available at website https://crop-leaderboard.github.io. | Fan Wu, Linyi Li, Zijian Huang, Yevgeniy Vorobeychik, Ding Zhao, Bo Li | null | null | 2,022 | iclr |
Patch-Fool: Are Vision Transformers Always Robust Against Adversarial Perturbations? | null | Vision transformers (ViTs) have recently set off a new wave in neural architecture design thanks to their record-breaking performance in various vision tasks. In parallel, to fulfill the goal of deploying ViTs into real-world vision applications, their robustness against potential malicious attacks has gained increasing attention. In particular, recent works show that ViTs are more robust against adversarial attacks as compared with convolutional neural networks (CNNs), and conjecture that this is because ViTs focus more on capturing global interactions among different input/feature patches, leading to their improved robustness to local perturbations imposed by adversarial attacks. In this work, we ask an intriguing question: "Under what kinds of perturbations do ViTs become more vulnerable learners compared to CNNs?" Driven by this question, we first conduct a comprehensive experiment regarding the robustness of both ViTs and CNNs under various existing adversarial attacks to understand the underlying reason favoring their robustness. Based on the drawn insights, we then propose a dedicated attack framework, dubbed Patch-Fool, that fools the self-attention mechanism by attacking its basic component (i.e., a single patch) with a series of attention-aware optimization techniques. Interestingly, our Patch-Fool framework shows for the first time that ViTs are not necessarily more robust than CNNs against adversarial perturbations. In particular, we find that ViTs are more vulnerable learners compared with CNNs against our Patch-Fool attack which is consistent across extensive experiments, and the observations from Sparse/Mild Patch-Fool, two variants of Patch-Fool, indicate an intriguing insight that the perturbation density and strength on each patch seem to be the key factors that influence the robustness ranking between ViTs and CNNs. It can be expected that our Patch-Fool framework will shed light on both future architecture designs and training schemes for robustifying ViTs towards their real-world deployment. Our codes are available at https://github.com/RICE-EIC/Patch-Fool. | Yonggan Fu, Shunyao Zhang, Shang Wu, Cheng Wan, Yingyan Lin | null | null | 2,022 | iclr |
Anisotropic Random Feature Regression in High Dimensions | null | In contrast to standard statistical wisdom, modern learning algorithms typically find their best performance in the overparameterized regime in which the model has many more parameters than needed to fit the training data. A growing number of recent works have shown that random feature models can offer a detailed theoretical explanation for this unexpected behavior, but typically these analyses have utilized isotropic distributional assumptions on the underlying data generation process, thereby failing to provide a realistic characterization of real-world models that are designed to identify and harness the structure in natural data. In this work, we examine the high-dimensional asymptotics of random feature regression in the presence of structured data, allowing for arbitrary input correlations and arbitrary alignment between the data and the weights of the target function. We define a partial order on the space of weight-data alignments and prove that generalization performance improves in response to stronger alignment. We also clarify several previous observations in the literature by distinguishing the behavior of the sample-wise and parameter-wise learning curves, finding that sample-wise multiple descent can occur at scales dictated by the eigenstructure of the data covariance, but that parameter-wise multiple descent is limited to double descent, although strong anisotropy can induce additional signatures such as wide plateaus and steep cliffs. Finally, these signatures are related to phase transitions in the spectrum of the feature kernel matrix, and unlike the double descent peak, persist even under optimal regularization. | Gabriel Mel, Jeffrey Pennington | null | null | 2,022 | iclr |
Which model to trust: assessing the influence of models on the performance of reinforcement learning algorithms for continuous control tasks | null | The need for algorithms able to solve Reinforcement Learning (RL) problems with few trials has motivated the advent of model-based RL methods. The reported performance of model-based algorithms has dramatically increased within recent years. However, it is not clear how much of the recent progress is due to improved algorithms or due to improved models. While different modeling options are available to choose from when applying a model-based approach, the distinguishing traits and particular strengths of different models are not clear. The main contribution of this work lies precisely in assessing the model influence on the performance of RL algorithms. A set of commonly adopted models is established for the purpose of model comparison. These include Neural Networks (NNs), ensembles of NNs, two different approximations of Bayesian NNs (BNNs), that is, the Concrete Dropout NN and the Anchored Ensembling, and Gaussian Processes (GPs). The model comparison is evaluated on a suite of continuous control benchmarking tasks. Our results reveal that significant differences in model performance do exist. The Concrete Dropout NN reports persistently superior performance. We summarize these differences for the benefit of the modeler and suggest that the model choice is tailored to the standards required by each specific application. | Giacomo Arcieri, David Wölfle, Eleni Chatzi | null | null | 2,022 | iclr |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.