_id
stringlengths 40
40
| text
stringlengths 0
10k
|
|---|---|
3c5c3e264e238fe1b76cfe528a0ef26b97f7309b | This paper proposes an edge-directed interpolation algorithm for natural images. The basic idea is to first estimate local covariance coefficients from a low-resolution image and then use these covariance estimates to adapt the interpolation at a higher resolution based on the geometric duality between the low-resolution covariance and the high-resolution covariance. The edge-directed property of covariance-based adaptation attributes to its capability of tuning the interpolation coefficients to match an arbitrarily oriented step edge. A hybrid approach of switching between bilinear interpolation and covariance-based adaptive interpolation is proposed to reduce the overall computational complexity. Two important applications of the new interpolation algorithm are studied: resolution enhancement of grayscale images and reconstruction of color images from CCD samples. Simulation results demonstrate that our new interpolation algorithm substantially improves the subjective quality of the interpolated images over conventional linear interpolation. |
77c512cbb832436e1a35ad434e6bb3d763799763 | Werner Reichardt Centre for Integrative Neuroscience and Institute of Theoretical Physics, University of Tübingen, Germany Bernstein Center for Computational Neuroscience, Tübingen, Germany Graduate School for Neural Information Processing, Tübingen, Germany Max Planck Institute for Biological Cybernetics, Tübingen, Germany Department of Neuroscience, Baylor College of Medicine, Houston, TX, USA ∗To whom correspondence should be addressed; E-mail: [email protected] |
7c14de73292d0398b73638f352f53c8b410049c1 | Purpose – The purpose of this paper is to develop a model that can be used to predict which small to medium-sized enterprises (SMEs) are more likely to become adopters of enterprise systems (ERP, CRM, SCM and e-procurement). Design/methodology/approach – Direct interviews were used to collect data from a random sample of SMEs located in the Northwest of England. Using logistic regression, 102 responses were analysed. Findings – The results reveal that the factors influencing SMEs’ adoption of enterprise systems are different from the factors influencing SMEs’ adoption of other previously studied information systems (IS) innovations. SMEs were found to be more influenced by technological and organisational factors than environmental factors. Moreover, the results indicate that firms with a greater perceived relative advantage, a greater ability to experiment with these systems before adoption, greater top management support, greater organisational readiness and a larger size are predicted to become adopters of enterprise systems. Research limitations/implications – Although this study focused on the factors that influence SMEs’ adoption of a set of enterprise systems (i.e. ERP, CRM, SCM and e-procurement), it fails to differentiate between factors that influence each of these systems. Practical implications – The model can be used to assist software vendors not only to develop marketing strategies that can target potential adopters, but also to develop strategies to increase the adoption of ES among SMEs. Originality/value – The paper contributes to the continuing research in IS innovations adoption/diffusion in the small business context. |
7af2c4d2f4927be654292a55f4abb04c63c76ee3 | |
ea39812bb04923faff76ba567649aceaf19a660e | Human gait is one of the most important biometric which has so far been neglected for use in medical diagnostics. In this paper, we make a feasibility study on human gait acquired from a wearable sensor based biometric suit called as Intelligent Gait Oscillation Detector (IGOD). This suit measures simultaneous gait oscillation from eight major joints (two knees, two hips, two elbows and two shoulders) of a human body. Techniques for analyzing and understanding the human gait patterns were developed. Variance in the gait oscillation was studied with respect to gait speed varying from 3km/hr to 5km/hr. Gender variance (male/female) gait oscillation has also been studied. A comprehensive analysis on human gait affected by knee joint movement and hip joint oscillation has been addressed with the arm swing effects. This analysis will provide us with an insight on human bipedal locomotion and its stability. We plan to create a repository of human gait oscillations which could extensively be analyzed for person identification and detecting walking problems in patients, which is detection of disease in the medical field. |
f2c4082faeff5d63b0144ef371c8964621ee33bf | Braincomputer interfaces (BCIs) allow their users to communicate or control external devices using brain signals rather than the brain's normal output pathways of peripheral nerves and muscles. Motivated by the hope of restoring independence to severely disabled individuals and by interest in further extending human control of external systems, researchers from many fields are engaged in this challenging new work. BCI research and development has grown explosively over the past two decades. Efforts have begun recently to provide laboratory-validated BCI systems to severely disabled individuals for real-world applications. In this paper, we discuss the current status and future prospects of BCI technology and its clinical applications. We will define BCI, review the BCI-relevant signals from the human brain, and describe the functional components of BCIs. We will also review current clinical applications of BCI technology and identify potential users and potential applications. Lastly, we will discuss current limitations of BCI technology, impediments to its widespread clinical use, and expectations for the future. |
b452a829d69fb1e265cf9277ff669bbc2fa8859b | Existing neural machine translation (NMT) models generally translate sentences in isolation, missing the opportunity to take advantage of document-level information. In this work, we propose to augment NMT models with a very light-weight cache-like memory network, which stores recent hidden representations as translation history. The probability distribution over generated words is updated online depending on the translation history retrieved from the memory, endowing NMT models with the capability to dynamically adapt over time. Experiments on multiple domains with different topics and styles show the effectiveness of the proposed approach with negligible impact on the computational cost. |
6a821cb17b30c26218e3eb5c20d609dc04a47bcb | Sparsity helps reduce the computational complexity of deep neural networks by skipping zeros. Taking advantage of sparsity is listed as a high priority in next generation DNN accelerators such as TPU[1]. The structure of sparsity, i.e., the granularity of pruning, affects the efficiency of hardware accelerator design as well as the prediction accuracy. Coarse-grained pruning creates regular sparsity patterns, making it more amenable for hardware acceleration but more challenging to maintain the same accuracy. In this paper we quantitatively measure the trade-off between sparsity regularity and prediction accuracy, providing insights in how to maintain accuracy while having more a more structured sparsity pattern. Our experimental results show that coarse-grained pruning can achieve a sparsity ratio similar to unstructured pruning without loss of accuracy. Moreover, due to the index saving effect, coarse-grained pruning is able to obtain a better compression ratio than fine-grained sparsity at the same accuracy threshold. Based on the recent sparse convolutional neural network accelerator (SCNN), our experiments further demonstrate that coarse-grained sparsity saves∼ 2× the memory references compared to fine-grained sparsity. Since memory reference is more than two orders of magnitude more expensive than arithmetic operations, the regularity of sparse structure leads to more efficient hardware design. |
0651b333c2669227b0cc42de403268a4546ece70 | Countless learning tasks require dealing with sequential data. Image captioning, speech synthesis, and music generation all require that a model produce outputs that are sequences. In other domains, such as time series prediction, video analysis, and musical information retrieval, a model must learn from inputs that are sequences. Interactive tasks, such as translating natural language, engaging in dialogue, and controlling a robot, often demand both capabilities. Recurrent neural networks (RNNs) are connectionist models that capture the dynamics of sequences via cycles in the network of nodes. Unlike standard feedforward neural networks, recurrent networks retain a state that can represent information from an arbitrarily long context window. Although recurrent neural networks have traditionally been difficult to train, and often contain millions of parameters, recent advances in network architectures, optimization techniques, and parallel computation have enabled successful large-scale learning with them. In recent years, systems based on long short-term memory (LSTM) and bidirectional (BRNN) architectures have demonstrated ground-breaking performance on tasks as varied as image captioning, language translation, and handwriting recognition. In this survey, we review and synthesize the research that over the past three decades first yielded and then made practical these powerful learning models. When appropriate, we reconcile conflicting notation and nomenclature. Our goal is to provide a selfcontained explication of the state of the art together with a historical perspective and references to primary research. |
db10480323bd2b9aeb58268678c204053722d6ad | |
168a89ce530c63720da844a30f5fce0c8f00fe8b | We investigate algorithms for evaluating sliding window joins over pairs of unbounded streams. We introduce a unittime-basis cost model to analyze the expected performance of these algorithms. Using this cost model, we propose strategies for maximizing the efficiency of processing joins in three scenarios. First, we consider the case where one stream is much faster than the other. We show that asymmetric combinations of join algorithms, (e.g., hash join on one input, nested-loops join on the other) can outperform symmetric join algorithm implementations. Second, we investigate the case where system resources are insufficient to keep up with the input streams. We show that we can maximize the number of join result tuples produced in this case by properly allocating computing resources across the two input streams. Finally, we investigate strategies for maximizing the number of result tuples produced when memory is limited, and show that proper memory allocation across the two input streams can result in significantly lower resource usage and/or more result tuples produced. |
39348c10c90be968357e2a6b65d5e0e479307735 | The Internet has become a rich and large repository of information about us as individuals. Anything from the links and text on a user's homepage to the mailing lists the user subscribes to are reflections of social interactions a user has in the real world. In this paper we devise techniques to mine this information in order to predict relationships between individuals. Further we show that some pieces of information are better indicators of social connections than others, and that these indicators vary between user populations and provide a glimpse into the social lives of individuals in different communities. Our techniques provide potential applications in automatically inferring real-world connections and discovering, labeling, and characterizing communities. |
d729b3a3f519a7b095891ed2b6aab459f6e121a9 | This paper studies the electrostatic discharge (ESD)-protected RF low-noise amplifiers (LNAs) in 65-nm CMOS technology. Three different ESD designs, including double-diode, modified silicon-controlled rectifier (SCR), and modified-SCR with double-diode configurations, are employed to realize ESD-protected LNAs at 5.8 GHz. By using the modified-SCR in conjunction with double-diode, a 5.8-GHz LNA with multiple ESD current paths demonstrates a 4.3-A transmission line pulse (TLP) failure level, corresponding to a ~ 6.5-kV Human-Body-Mode (HBM) ESD protection level. Under a supply voltage of 1.2 V and a drain current of 6.5 mA, the proposed ESD-protected LNA demonstrates a noise figure of 2.57 dB with an associated power gain of 16.7 dB. The input third-order intercept point (IIP3) is - 11 dBm, the input and output return losses are greater than 15.9 and 20 dB, respectively. |
b1397c9085361f308bd70793fc2427a4416973d7 | This paper describes a probabilistic approach to the problem of recognizing places based on their appearance. The system we present is not limited to localization, but can determine that a new observation comes from a previously unseen place, and so augment its map. Effectively this is a SLAM system in the space of appearance. Our probabilistic approach allows us to explicitly account for perceptual aliasing in the environment—identical but indistinctive observations receive a low probability of having come from the same place. We achieve this by learning a generative model of place appearance. By partitioning the learning problem into two parts, new place models can be learned online from only a single observation of a place. The algorithm complexity is linear in the number of places in the map, and is particularly suitable for online loop closure detection in mo- |
988058ab8dfcb27e9566c6bcef398a4407b1ea04 | |
b320b4b23f708344b7bc4af20fdb37e56543d1a2 | We present a simple method to incorporate syntactic information about the target language in a neural machine translation system by translating into linearized, lexicalized constituency trees. Experiments on the WMT16 German-English news translation task shown improved BLEU scores when compared to a syntax-agnostic NMT baseline trained on the same dataset. An analysis of the translations from the syntax-aware system shows that it performs more reordering during translation in comparison to the baseline. A smallscale human evaluation also showed an advantage to the syntax-aware system. ”accepted...” header |
3d809bb3b414a8ee58492e7ea775d6631ea05e91 | Cantonese is an important dialect in some regions of Southern China. Local online users often represent their opinions and experiences on the web with written Cantonese. Although the information in those reviews is valuable to potential consumers and sellers, the huge amount of web reviews make it difficult to give an unbiased evaluation to a product and the Cantonese reviews are unintelligible for Mandarin Chinese speakers. In this paper, standard machine learning techniques naive Bayes and SVM are incorporated into the domain of online Cantonese-written restaurant reviews to automatically classify user reviews as positive or negative. The effects of feature presentations and feature sizes on classification performance are discussed. We find that accuracy is influenced by interaction between the classification models and the feature options. The naive Bayes classifier achieves as well as or better accuracy than SVM. Character-based bigrams are proved better features than unigrams and trigrams in capturing Cantonese sentiment orientation. 2010 Elsevier Ltd. All rights reserved. |
771636b26260fac6d215df5e76c9ce72c346ba88 | |
025720574ef67672c44ba9e7065a83a5d6075c36 | We use multilayer Long Short Term Memory (LSTM) networks to learn representations of video sequences. Our model uses an encoder LSTM to map an input sequence into a fixed length representation. This representation is decoded using single or multiple decoder LSTMs to perform different tasks, such as reconstructing the input sequence, or predicting the future sequence. We experiment with two kinds of input sequences – patches of image pixels and high-level representations (“percepts”) of video frames extracted using a pretrained convolutional net. We explore different design choices such as whether the decoder LSTMs should condition on the generated output. We analyze the outputs of the model qualitatively to see how well the model can extrapolate the learned video representation into the future and into the past. We try to visualize and interpret the learned features. We stress test the model by running it on longer time scales and on out-of-domain data. We further evaluate the representations by finetuning them for a supervised learning problem – human action recognition on the UCF-101 and HMDB-51 datasets. We show that the representations help improve classification accuracy, especially when there are only a few training examples. Even models pretrained on unrelated datasets (300 hours of YouTube videos) can help action recognition performance. |
14316b885f65d2197ce8c6d4ab3ee61fdab052b8 | This publication contains reprint articles for which IEEE does not hold copyright. Full text is not available on IEEE Xplore for these articles. |
3068d1d6275933e7f4d332a2f2cf52543a4f0615 | A homomorphic, or incremental, multiset hash function, associates a hash value to arbitrary collections of objects (with possible repetitions) in such a way that the hash of the union of two collections is easy to compute from the hashes of the two collections themselves: it is simply their sum under a suitable group operation. In particular, hash values of large collections can be computed incrementally and/or in parallel. Homomorphic hashing is thus a very useful primitive with applications ranging from database integrity verification to streaming set/multiset comparison and network coding. Unfortunately, constructions of homomorphic hash functions in the literature are hampered by two main drawbacks: they tend to be much longer than usual hash functions at the same security level (e.g. to achieve a collision resistance of 2, they are several thousand bits long, as opposed to 256 bits for usual hash functions), and they are also quite slow. In this paper, we introduce the Elliptic Curve Multiset Hash (ECMH), which combines a usual bit string-valued hash function like BLAKE2 with an efficient encoding into binary elliptic curves to overcome both difficulties. On the one hand, the size of ECMH digests is essentially optimal: 2m-bit hash values provideO(2) collision resistance. On the other hand, we demonstrate a highly-efficient software implementation of ECMH, which our thorough empirical evaluation shows to be capable of processing over 3 million set elements per second on a 4 GHz Intel Haswell machine at the 128-bit security level— many times faster than previous practical methods. While incremental hashing based on elliptic curves has been considered previously [1], the proposed method was less efficient, susceptible to timing attacks, and potentially patent-encumbered [2], and no practical implementation was demonstrated. |
fbcd758ecc083037cd035c8ed0c26798ce62f15e | We present a method for detecting exploitations of vulnerabilities in privileged programs by monitoring their execution using audit trials, where the monitoring is with respect to specifications of the security-relevant behavior of the programs. Our work is motivated by the intrusion detection paradigm, but is a n attempt to avoid a d hoc approaches to codifying misuse behavior. Our approach as based on the observation that although privileged programs can be exploited (due to errors) to cause security compromise in systems because of the privileges accorded to them, the intended behavior of privileged programs is, of course, limited a n d benign. The key, then i s to specify the intended behavior (Le., the program policy) and to detect any action by privileged program that is outside the intended behavior a n d that imperils security. We describe a program policy specification language, which is based on simple predicate logic and regular expressions. I n addition, we present specifications of privileged programs in Unix, and Q prototype execution monitor for analyzing audit trails with respect to these specifications. The program policies are surprisingly concise and clear, a n d an addition, capable of detecting exploitations of known vulnerabilities in these programs. Although our work has been motivated by the known vulnerabilities in Unix, we believe that by tightly restricting the behavior of all privileged programs, exploitations of unknown vulnerabilities can be detected. A s Q check on the specifications, work is in progress on verifying them with respect to a n abstract security policy. *This work is funded in part by the National Security Agency University Research Program under Contract No. DOD-MDA904-93-C4083 and by ARPA under Contract No. USNN00014-94-1-0065. |
cb2111cb362f566be61c75ada38af53ecf95a1d3 | |
564264f0ce2b26fd80320e4f71b70ee8c67602ef | In this paper, we address the problem of unsupervised domain transfer learning in which no labels are available in the target domain. We use a transformation matrix to transfer both the source and target data to a common subspace, where each target sample can be represented by a combination of source samples such that the samples from different domains can be well interlaced. In this way, the discrepancy of the source and target domains is reduced. By imposing joint low-rank and sparse constraints on the reconstruction coefficient matrix, the global and local structures of data can be preserved. To enlarge the margins between different classes as much as possible and provide more freedom to diminish the discrepancy, a flexible linear classifier (projection) is obtained by learning a non-negative label relaxation matrix that allows the strict binary label matrix to relax into a slack variable matrix. Our method can avoid a potentially negative transfer by using a sparse matrix to model the noise and, thus, is more robust to different types of noise. We formulate our problem as a constrained low-rankness and sparsity minimization problem and solve it by the inexact augmented Lagrange multiplier method. Extensive experiments on various visual domain adaptation tasks show the superiority of the proposed method over the state-of-the art methods. The MATLAB code of our method will be publicly available at http://www.yongxu.org/lunwen.html. |
13645dd03627503fd860a2ba73189e92393a67e3 | In this paper we develop new Newton and conjugate gradient algorithms on the Grassmann and Stiefel manifolds. These manifolds represent the constraints that arise in such areas as the symmetric eigenvalue problem, nonlinear eigenvalue problems, electronic structures computations, and signal processing. In addition to the new algorithms, we show how the geometrical framework gives penetrating new insights allowing us to create, understand, and compare algorithms. The theory proposed here provides a taxonomy for numerical linear algebra algorithms that provide a top level mathematical view of previously unrelated algorithms. It is our hope that developers of new algorithms and perturbation theories will benefit from the theory, methods, and examples in this paper. |
38d555bfe13b61e838364016219c7e42fb5dc919 | Purpose – The purpose of this paper is to propose a business process management (BPM) governance model that sets BPM decision making, along with roles and responsibilities. The setting context of the study is a government-owned corporation operating in Australia. Design/methodology/approach – A qualitative case study examined and analysed organisational documents using a content analysis approach. Results of document analysis are used to inform a series of in-depth interviews of key stakeholders in the organisation. Interviews are analysed using a constant comparison method to derive themes and build categories of description. Findings – A BPM governance model is proposed. Results of thematic analysis are interpreted against the framework of the BPM governance model, leading to findings that include implications for theory and practice. Practical implications – In practical terms, the research shows how BPM practice can be aligned and integrated with the corporate governance and management systems in the selected case study organisation. Originality/value – Despite research identifying the importance of governance, along with associated capabilities, there has been little progress on how the abovementioned capabilities can be effectively deployed across an organisation. This paper addresses a gap in the literature relating to how to deploy BPM governance in an organisation. |
268d3f28ae2295b9d2bf6fef2aa27faf9048a86c | We propose to exploit the potential of multiple word clusterings in class-based recurrent neural network (RNN) language models for ensemble RNN language modeling. By varying the clustering criteria and the space of word embedding, different word clusterings are obtained to define different word/class factorizations. For each such word/class factorization, several base RNNLMs are learned, and the word prediction probabilities of the base RNNLMs are then combined to form an ensemble prediction. We use a greedy backward model selection procedure to select a subset of models and combine these models for word prediction. The proposed ensemble language modeling method has been evaluated on Penn Treebank test set as well as Wall Street Journal (WSJ) Eval 92 and 93 test sets, where it improved test set perplexity and word error rate over the state-of-the-art single RNNLMs as well as multiple RNNLMs produced by varying RNN learning conditions. |
05a958194f1756fb91ddd3e2cd5794f9b0c312ce | Multilevel inverters have created a new wave of interest in industry and research. While the classical topologies have proved to be a viable alternative in a wide range of high-power medium-voltage applications, there has been an active interest in the evolution of newer topologies. Reduction in overall part count as compared to the classical topologies has been an important objective in the recently introduced topologies. In this paper, some of the recently proposed multilevel inverter topologies with reduced power switch count are reviewed and analyzed. The paper will serve as an introduction and an update to these topologies, both in terms of the qualitative and quantitative parameters. Also, it takes into account the challenges which arise when an attempt is made to reduce the device count. Based on a detailed comparison of these topologies as presented in this paper, appropriate multilevel solution can be arrived at for a given application. |
79357470d76ae7aeb4f6e39efd7c3936d615caa4 | This paper1 presents a linear optical receiver designed using a 28nm CMOS technology suitable for 20Gbaud/s (40Gb/s) PAM-4 modulation. The optical receiver consists of a transimpedance amplifier (gain adjustable from 40dBΩ to 56dBO) followed by a variable gain amplifier (gain adjustable from 6dB to 17dB). Capacitive peaking is used to achieve a bandwidth of ~10GHz, thus avoiding the use of on-chip inductors which require large die area. A robust automatic gain control loop is used to ensure a constant differential output voltage swing of ~100mV for an input dynamic range of 20μA to 500μA (peak current). Over this same range, high linearity (total harmonic distortion less than 5%, 250MHz sinewave, 10harmonics taken into account) is obtained. The rms input referred noise current (integrated from 10MHz to 20GHz) was 2.5μArms. The linear optical receiver consumes 56mW from a 1.5V supply voltage. |
2c361ef5db3231d34656dd86d9b288397f0b929e | A framework is developed to explore the connection between e ective optimization algorithms and the problems they are solving A number of no free lunch NFL theorems are presented that establish that for any algorithm any elevated performance over one class of problems is exactly paid for in performance over another class These theorems result in a geometric interpretation of what it means for an algorithm to be well suited to an optimization problem Applications of the NFL theorems to information theoretic aspects of optimization and benchmark measures of performance are also presented Other issues addressed are time varying optimization problems and a priori head to head minimax distinctions between optimization algorithms distinctions that can obtain despite the NFL theorems enforcing of a type of uniformity over all algorithms |
bc87585b4fc874a29bd1d9a031dce1807b2ba0e8 | Given a topic of interest, a contrastive theme is a group of opposing pairs of viewpoints. We address the task of summarizing contrastive themes: given a set of opinionated documents, select meaningful sentences to represent contrastive themes present in those documents. Several factors make this a challenging problem: unknown numbers of topics, unknown relationships among topics, and the extraction of comparative sentences. Our approach has three core ingredients: contrastive theme modeling, diverse theme extraction, and contrastive theme summarization. Specifically, we present a hierarchical non-parametric model to describe hierarchical relations among topics; this model is used to infer threads of topics as themes from the nested Chinese restaurant process. We enhance the diversity of themes by using structured determinantal point processes for selecting a set of diverse themes with high quality. Finally, we pair contrastive themes and employ an iterative optimization algorithm to select sentences, explicitly considering contrast, relevance, and diversity. Experiments on three datasets demonstrate the effectiveness of our method. |
147fe6bfc76f30ccacc3620662511e452bc395f6 | Face recognition presents a challenging problem in the field of image analysis and computer vision, and as such has received a great deal of attention over the last few years because of its many applications in various domains. Face recognition techniques can be broadly divided into three categories based on the face data acquisition methodology: methods that operate on intensity images; those that deal with video sequences; and those that require other sensory data such as 3D information or infra-red imagery. In this paper, an overview of some of the well-known methods in each of these categories is provided and some of the benefits and drawbacks of the schemes mentioned therein are examined. Furthermore, a discussion outlining the incentive for using face recognition, the applications of this technology, and some of the difficulties plaguing current systems with regard to this task has also been provided. This paper also mentions some of the most recent algorithms developed for this purpose and attempts to give an idea of the state of the art of face recognition technology. |
26d172f0a4d7e903ce388f3159059f9c5463e5c5 | The extraction of image features is one of the fundamental tasks in image recognition . Up until now, there have been several kinds of features to be used for the purpose of image recognition as follows: (1) visual features ; (2) statistical features of pixel ; (3) transform coefficient features . In addition, there is another kind of feature which the author believes is very useful, i.e . (4) algebraic features which represent intrinsic attributions of an image . Singular Values (SV) of image are this kind of feature . In this paper, we prove that SV feature vector has some important properties of algebraic and geometric invariance, and insensitiveness to noise . These properties are very useful for the description and recognition of images . As an example, SV feature vector is used for the problem of recognizing human facial images . In this paper, using SV feature vector samples of facial images, a normal pattern Bayes classification model based on Sammon s optimal descriminant plane is constructed . The experimental result shows that SV feature vector has good performance of class separation . Image recognition Algebraic feature extraction Singular value feature Facial image recognition Discrinlinant vector Dimensionality reduction |
55206f0b5f57ce17358999145506cd01e570358c | Stochastic modelling of non-stationary vector timeseries based on HMMs has been very successful for speech applications [5]. Recently it has been applied to a range of image recognition problems [7, 9]. Previously reported work [6] has investigated the use of HMMs to model human faces for identi cation purposes. Faces can be intuitively divided into regions such as the mouth, eyes, nose, etc., and these regions can be associated with the states of an HMM. The identi cation performance of a top-bottom HMM compares favourably with some of the well-known algorithms, for example eigenfaces as detailed in [8]. However, the HMM parameterisation in the work presented so far was arrived at by subjective intuition. This paper presents experimental results which show how identi cation rates vary with HMM parameters, and which indicate the most sensible choice of parameters. The paper is organised as follows: section 2 gives an overview of the HMM-based approach; section 3 details the training and recognition processes; section 4 describes the experimental setup; section 5 presents the identi cation results; section 6 concludes the paper. |
5985014dda6d502469614aae17349b4d08f9f74c | -This paper evaluates the performance both of some texture measures which have been successfully used in various applications and of some new promising approaches proposed recently. For classification a method based on Kullback discrimination of sample and prototype distributions is used. The classification results for single features with one-dimensional feature value distributions and for pairs of complementary features with two-dimensional distributions are presented Texture analysis Classification Feature distribution Kullback discriminant Performance evaluation Brodatz textures l. I N T R O D U C T I O N Texture is an important characteristic for the analysis of many types of images. A wide variety of measures for discriminating textures have been proposed, c1'2~ Comparative studies to evaluate the performance of some texture measures have been carried out by Weszka et al., ~3~ Du Bufet al. ~4~ and Ohanian and Dubes, ~s) for |
1270044a3fa1a469ec2f4f3bd364754f58a1cb56 | This paper presents a novel method to model and recognize human faces in video sequences. Each registered person is represented by a low-dimensional appearance manifold in the ambient image space. The complex nonlinear appearance manifold expressed as a collection of subsets (named pose manifolds), and the connectivity among them. Each pose manifold is approximated by an affine plane. To construct this representation, exemplars are sampled from videos, and these exemplars are clustered with a K-means algorithm; each cluster is represented as a plane computed through principal component analysis (PCA). The connectivity between the pose manifolds encodes the transition probability between images in each of the pose manifold and is learned from a training video sequences. A maximum a posteriori formulation is presented for face recognition in test video sequences by integrating the likelihoo d that the input image comes from a particular pose manifold and the transition probability to this pose manifold from th e previous frame. To recognize faces with partial occlusion, we introduce a weight mask into the process. Extensive experiments demonstrate that the proposed algorithm outperforms existing frame-based face recognition methods with temporal voting schemes. |
358e2ae243cb022938ae3d40ea0ac112319a6325 | |
ad9cbf31a1cd6a71e773a0d3e93c489304327174 | This paper describes a powered leg orthosis for gait rehabilitation of patients with walking disabilities. The paper proposes controllers which can apply suitable forces on the leg so that it moves on a desired trajectory. The description of the controllers, simulations and experimental results with the powered orthosis are presented in the paper. Currently, experiments have been performed with a dummy leg in the orthosis. In the coming months, this powered orthosis will be used on healthy subjects and stroke patients. |
853ac9d5ae2662b8e33946d106b261005e391fed | The randomness and richness present in the iris texture make the 2D Gabor filter bank analysis a suitable technique to be used for iris recognition systems. To accurately characterize complex texture structures using 2D Gabor filters it is necessary to use multiple sets of parameters of this type of filters. This paper proposes a technique of optimizing multiple sets of 2D Gabor filter parameters to gradually enhance the accuracy of an iris recognition system. The proposed methodology is suitable to be applied on both near infrared and visible spectrum iris images. To illustrate the efficiency of the filter bank design technique, UBIRISv1 database was used for benchmarking. |
050eda213ce29da7212db4e85f948b812a215660 | We propose a modeland exemplar-based approach for face recognition. This problem has been previously tackled using either models or exemplars, with limited success. Our idea uses models to synthesize many more exemplars, which are then used in the learning stage of a face recognition system. To demonstrate this, we develop a statistical shape-fromshading model to recover face shape from a single image, and to synthesize the same face under new illumination. We then use this to build a simple and fast classifier that was not possible before because of a lack of training data. |
195e55c90fd109642116ee51f7205c106f341111 | |
b05fdba8f447b37d7fa6fdd63d23c70b2f4ee01b | In this paper, we present the first deep learning approach to aspect extraction in opinion mining. Aspect extraction is a subtask of sentiment analysis that consists in identifying opinion targets in opinionated text, i.e., in detecting the specific aspects of a product or service the opinion holder is either praising or complaining about. We used a 7-layer deep convolutional neural network to tag each word in opinionated sentences as either aspect or non-aspect word. We also developed a set of linguistic patterns for the same purpose and combined them with the neural network. The resulting ensemble classifier, coupled with a word-embedding model for sentiment analysis, allowed our approach to obtain significantly better accuracy than state-of-the-art methods. |
d9c797a74f2f716379f388d5c57e437ecd68759c | |
6fdbbefe05648f6c0f027428ccff248b174798d5 | In this paper, we discuss a significant open problem in mobile robotics: simultaneous map building and localization, which we define as long-term globally referenced position estimation without a priori information. This problem is difficult because of the following paradox: to move precisely, a mobile robot must have an accurate environment map; however, to build an accurate map, the mobile robot’s sensing locations must be known precisely. In this way, simultaneous map building and localization can be seen to present a question of “which came first, the chicken or the egg?” (The map or the motion?) When using ultrasonic sensing, to overcome this issue we equip the vehicle with multiple servo-mounted sonar sensors, to provide a means in which a subset of environment features can be precisely learned from the robot’s initial location and subsequently tracked to provide precise positioning. |
efbc200feab74e5087c4005d8759e5dadb3a3077 | Generic generation and manipulation of text is challenging and has limited success compared to recent deep generative modeling in visual domain. This paper aims at generating plausible natural language sentences, whose attributes are dynamically controlled by learning disentangled latent representations with designated semantics. We propose a new neural generative model which combines variational auto-encoders and holistic attribute discriminators for effective imposition of semantic structures. With differentiable approximation to discrete text samples, explicit constraints on independent attribute controls, and efficient collaborative learning of generator and discriminators, our model learns highly interpretable representations from even only word annotations, and produces realistic sentences with desired attributes. Quantitative evaluation validates the accuracy of sentence and attribute generation. |
24aed1b7277dfb2c2a6515a1be82d30cc8aa85cc | We address the challenge of sentiment analysis from visual content. In contrast to existing methods which infer sentiment or emotion directly from visual low-level features, we propose a novel approach based on understanding of the visual concepts that are strongly related to sentiments. Our key contribution is two-fold: first, we present a method built upon psychological theories and web mining to automatically construct a large-scale Visual Sentiment Ontology (VSO) consisting of more than 3,000 Adjective Noun Pairs (ANP). Second, we propose SentiBank, a novel visual concept detector library that can be used to detect the presence of 1,200 ANPs in an image. The VSO and SentiBank are distinct from existing work and will open a gate towards various applications enabled by automatic sentiment analysis. Experiments on detecting sentiment of image tweets demonstrate significant improvement in detection accuracy when comparing the proposed SentiBank based predictors with the text-based approaches. The effort also leads to a large publicly available resource consisting of a visual sentiment ontology, a large detector library, and the training/testing benchmark for visual sentiment analysis. |
9af5c320f1ab4e881c1aa4e35d7c7f10d5d2405d | |
b6bbb228300c72f141a2f05702ddc7f8ab4a8297 | Information is the life blood of all modern organizations yet the news media continue to report stories of critical information loss. The purpose of information security is to protect valuable assets, such as information, hardware, software and people. The majority of information security specialists believe that promoting good end user behavior and constraining bad end user behavior is an important component of an effective Information Security Management System (ISMS). Implementing effective information security involves understanding security-related risk, then developing and implementing appropriate controls. In general the better employees are at applying the controls the more secure the organization will be, because even the best designed technical controls and procedures will be of limited value if the staff involved do not understand why they have been implemented and what they are accomplishing. Achieving the required level of understanding usually requires more than an annual awareness training initiative and represents a major challenge for most organizations. In fact, for many organizations it will involve a cultural change to ensure the integration of information security concepts into the organizational culture. |
0bd6442092bc4a9e0e77cd2f302f2db1a242e250 | Health monitoring systems based on Internet-of-things (IoT) have been recently introduced to improve the quality of health care services. However, the number of advanced IoT-based continuous glucose monitoring systems is small and the existing systems have several limitations. In this paper we study feasibility of invasive and continuous glucose monitoring (CGM) system utilizing IoT based approach. We designed an IoT-based system architecture from a sensor device to a back-end system for presenting real-time glucose, body temperature and contextual data (i.e. environmental temperature) in graphical and human-readable forms to end-users such as patients and doctors. In addition, nRF communication protocol is customized for suiting to the glucose monitoring system and achieving a high level of energy efficiency. Furthermore, we investigate energy consumption of the sensor device and design energy harvesting units for the device. Finally, the work provides many advanced services at a gateway level such as a push notification service for notifying patient and doctors in case of abnormal situations (i.e. too low or too high glucose level). The results show that our system is able to achieve continuous glucose monitoring remotely in real-time. In addition, the results reveal that a high level of energy efficiency can be achieved by applying the customized nRF component, the power management unit and the energy harvesting unit altogether in the sensor device. c © 2017 The Authors. Published by E sevier B.V. i ilit f t f re ce r ra hairs. |
0bfc3626485953e2d3f87854a00a50f88c62269d | Cellular networks are usually modeled by placing the base stations on a grid, with mobile users either randomly scattered or placed deterministically. These models have been used extensively but suffer from being both highly idealized and not very tractable, so complex system-level simulations are used to evaluate coverage/outage probability and rate. More tractable models have long been desirable. We develop new general models for the multi-cell signal-to-interference-plus-noise ratio (SINR) using stochastic geometry. Under very general assumptions, the resulting expressions for the downlink SINR CCDF (equivalent to the coverage probability) involve quickly computable integrals, and in some practical special cases can be simplified to common integrals (e.g., the Q-function) or even to simple closed-form expressions. We also derive the mean rate, and then the coverage gain (and mean rate loss) from static frequency reuse. We compare our coverage predictions to the grid model and an actual base station deployment, and observe that the proposed model is pessimistic (a lower bound on coverage) whereas the grid model is optimistic, and that both are about equally accurate. In addition to being more tractable, the proposed model may better capture the increasingly opportunistic and dense placement of base stations in future networks. |
54bef8bca4bef4a5cb597c11b9389496f40df35c | Punctuations are not available in automatic speech recognition outputs, which could create barriers to many subsequent text processing tasks. This paper proposes a novel method to predict punctuation symbols for the stream of words in transcribed speech texts. Our method jointly performs parsing and punctuation prediction by integrating a rich set of syntactic features when processing words from left to right. It can exploit a global view to capture long-range dependencies for punctuation prediction with linear complexity. The experimental results on the test data sets of IWSLT and TDT4 show that our method can achieve high-level performance in punctuation prediction over the stream of words in transcribed speech text. |
82e9a883f47380ce2c89ecbc57597efbdd120be1 | Radio-frequency identification technology, based on the reader/tag paradigm, is quickly permeating several aspects of everyday life. The electromagnetic research mainly concerns the design of tag antennas having high efficiency and small size, and suited to complex impedance matching to the embedded electronics. Starting from the available but fragmented open literature, this paper presents a homogeneous survey of relevant methodologies for the design of UHF passive tag antennas. Particular care is taken to illustrate, within a common framework, the basic concepts of the most-used design layouts. The design techniques are illustrated by means of many noncommercial examples. |
54a4c8051e655e3035f98bb9dd8876a6511517ff | This paper describes LIMSI’s submissions to the shared WMT’16 task “Translation of News”. We report results for Romanian-English in both directions, for English to Russian, as well as preliminary experiments on reordering to translate from English into German. Our submissions use mainly NCODE and MOSES along with continuous space models in a post-processing step. The main novelties of this year’s participation are the following: for the translation into Russian and Romanian, we have attempted to extend the output of the decoder with morphological variations and to use a CRF model to rescore this new search space; as for the translation into German, we have been experimenting with source-side pre-ordering based on a dependency structure allowing permutations in order to reproduce the tar- |
e092be3120ce51aaa35af673fabd07b6513e702e | Many of the problems arjd failures of Management Information Systems (MIS) and Management Science/Operations Research (MS/OR) projects have been attributed to organizational behavioral problems. The millions of dollars organizations spend on MIS and MS/OR development are of little benefit because systems continue to fail. Steps can be taken to understand and solve these behavioral probiems. This article argues that in most oases these behavioral problems are the result of inadequate designs. These bad designs are attributed to the way MIS systems designers view organizations, their members, and the function of an MIS within them. I.e., systems designers' frames of reference. These frames of reference cause faulty design choices and failures to perceive better design alternatives. Seven conditions are discussed which reflect current systems designers' points of view. The discussion of these conditions demonstrates the need to refrawe MIS design methodology within the Socio-Technicai Systems (STS) design approach and change systems designers' perspectives. The STS approach is introduced as a realistic view of organizations and a way to change them. This article is the first of two to appear in consecutive issues of the MIS Quarterly. The purpose of this first article is to demonstrate the need for the srs approaj:h. The second will present the basic concepts and principles of the STS methodology and how it can be utilized in the design of an MIS. |
26b6341330085c8588b0d6e3eaf34ab5a0f7ca53 | In this paper a new Voltage Controlled Ring Oscillator is presented. The proposed VCO uses partial positive feedback in its delay cell, allowing the circuit to operate with single two stages, achieving high speed with reduced power consumption. The new VCO has a wide range of operation frequencies (0.2 at 2.1 GHz), good linearity between the output frequency and the control voltage, phase noise of -90 dBc/Hz at 100 kHz offset and it only consumes 7.01 mW in its central frequency of 1.2 GHz using a 3.3 V power supply. The circuit was fabricated in a 0.35 /spl mu/m CMOS-AMS process and occupies an area of 67.5/spl times/77.5 /spl mu/m/sup 2/. |
4d4be6294e5b30cdf985fcc044f44ec9da495af3 | Computer-based sensors and actuators such as global positioning systems, machine vision, and laser-based sensors have progressively been incorporated into mobile robots with the aim of configuring autonomous systems capable of shifting operator activities in agricultural tasks. However, the incorporation of many electronic systems into a robot impairs its reliability and increases its cost. Hardware minimization, as well as software minimization and ease of integration, is essential to obtain feasible robotic systems. A step forward in the application of automatic equipment in agriculture is the use of fleets of robots, in which a number of specialized robots collaborate to accomplish one or several agricultural tasks. This paper strives to develop a system architecture for both individual robots and robots working in fleets to improve reliability, decrease complexity and costs, and permit the integration of software from different developers. Several solutions are studied, from a fully distributed to a whole integrated architecture in which a central computer runs all processes. This work also studies diverse topologies for controlling fleets of robots and advances other prospective topologies. The architecture presented in this paper is being successfully applied in the RHEA fleet, which comprises three ground mobile units based on a commercial tractor chassis. |
cbc5d3e04f80a07b49ac3fbdb41f4bd577664cdc | Realizing inter-satellite links is a must for ensuring the success of cubesat swarm missions. Nevertheless, it has hardly been considered until now. The communication systems for cubesats have to deal with a few peculiar demands regarding consumed power, geometry and throughput. Depending on the type of application, required data rates can go up to tens of megabits per second, while power consumption and physical size are limited by the platform. The proposed communication scheme will combine power-efficient modulation and channel coding with multiple access and spread spectrum techniques, enabling the deployment of multiple satellites. Apart from this, the antenna system has to be designed such that links can be established and maintained independent of the satellites' orientation. An electrically steerable radiation pattern is achieved by placing antennas on each face of the cube. Conformal beamforming provides the system with 5 dBi gain for any desired direction of transmission, eliminating the need for attitude control. Furthermore, using planar antennas reduces the complexity of the mechanical part as they require no deployment. |
029ec0d53785eaf719632f5aa67ae5c22689dc70 | A circularly polarized slot radiator in millimetre band is presented. Low reflection and good polarization performance is obtained in comparison with state of art slot elements. A high gain array prototype has been implemented in substrate integrated waveguides (SIW) technology, in order to check the new element performance. A cosecant amplitude coverage in the elevation plane, as well as a 6° tilted monopulse pattern in the azimuth plane are obtained. A gain peak value of 28.6 dBi and 79% of efficiency at 36.7 GHz have been obtained in the specified operation band (36.7 – 37 GHz) for a manufactured antenna prototype. Likewise, the top measured axial ratio is 1.95 dB at 36.85 GHz. |
e17879cf2bb858fbbc3a3aa2441287c53a0f684a | Trisomy X is a sex chromosome anomaly with a variable phenotype caused by the presence of an extra X chromosome in females (47,XXX instead of 46,XX). It is the most common female chromosomal abnormality, occurring in approximately 1 in 1,000 female births. As some individuals are only mildly affected or asymptomatic, it is estimated that only 10% of individuals with trisomy X are actually diagnosed. The most common physical features include tall stature, epicanthal folds, hypotonia and clinodactyly. Seizures, renal and genitourinary abnormalities, and premature ovarian failure (POF) can also be associated findings. Children with trisomy X have higher rates of motor and speech delays, with an increased risk of cognitive deficits and learning disabilities in the school-age years. Psychological features including attention deficits, mood disorders (anxiety and depression), and other psychological disorders are also more common than in the general population. Trisomy X most commonly occurs as a result of nondisjunction during meiosis, although postzygotic nondisjunction occurs in approximately 20% of cases. The risk of trisomy X increases with advanced maternal age. The phenotype in trisomy X is hypothesized to result from overexpression of genes that escape X-inactivation, but genotype-phenotype relationships remain to be defined. Diagnosis during the prenatal period by amniocentesis or chorionic villi sampling is common. Indications for postnatal diagnoses most commonly include developmental delays or hypotonia, learning disabilities, emotional or behavioral difficulties, or POF. Differential diagnosis prior to definitive karyotype results includes fragile X, tetrasomy X, pentasomy X, and Turner syndrome mosaicism. Genetic counseling is recommended. Patients diagnosed in the prenatal period should be followed closely for developmental delays so that early intervention therapies can be implemented as needed. School-age children and adolescents benefit from a psychological evaluation with an emphasis on identifying and developing an intervention plan for problems in cognitive/academic skills, language, and/or social-emotional development. Adolescents and adult women presenting with late menarche, menstrual irregularities, or fertility problems should be evaluated for POF. Patients should be referred to support organizations to receive individual and family support. The prognosis is variable, depending on the severity of the manifestations and on the quality and timing of treatment. |
5b42ed20a1a01cb8d097141303dfd8f7cf1ced10 | Shortest path queries (SPQ) are essential in many graph analysis and mining tasks. However, answering shortest path queries on-the-fly on large graphs is costly. To online answer shortest path queries, we may materialize and index shortest paths. However, a straightforward index of all shortest paths in a graph of N vertices takes O(N2) space. In this paper, we tackle the problem of indexing shortest paths and online answering shortest path queries. As many large real graphs are shown richly symmetric, the central idea of our approach is to use graph symmetry to reduce the index size while retaining the correctness and the efficiency of shortest path query answering. Technically, we develop a framework to index a large graph at the orbit level instead of the vertex level so that the number of breadth-first search trees materialized is reduced from O(N) to O(|Δ|), where |Δ| ≤ N is the number of orbits in the graph. We explore orbit adjacency and local symmetry to obtain compact breadth-first-search trees (compact BFS-trees). An extensive empirical study using both synthetic data and real data shows that compact BFS-trees can be built efficiently and the space cost can be reduced substantially. Moreover, online shortest path query answering can be achieved using compact BFS-trees. |
263c66b0c2dc996c46d11693ad3d6f8d3f7f3d3c | |
6bf1770a79309e4f05ef65e65e19f99c25974657 | What role will ubiquitous sensing play in our understanding and experience of ecology in the future? What opportunities are created by weaving a continuously sampling, geographically dense web of sensors into the natural environment, from the ground up? In this article, we explore these questions holistically, and present our work on an environmental sensor network designed to support a diverse array of applications, interpretations, and artistic expressions, from primary ecological research to musical composition. Over the past four years, we have been incorporating our ubiquitous sensing framework into the design and implementation of a large-scale wetland restoration, creating a broad canvas for creative exploration at the landscape scale. The projects we present here span the development and wide deployment of custom sensor node hardware, novel web services for providing real-time sensor data to end user applications, public-facing user interfaces for open-ended exploration of the data, as well as more radical UI modalities, through unmanned aerial vehicles, virtual and augmented reality, and wearable devices for sensory augmentation. From this work, we distill the Networked Sensory Landscape, a vision for the intersection of ubiquitous computing and environmental restoration. Sensor network technologies and novel approaches to interaction promise to reshape presence, opening up sensorial connections to ecological processes across spatial and temporal scales. |
c5cc6243f070d80f5edef24608694c39195e2d1a | The SQL Server 11 release (code named "Denali") introduces a new data warehouse query acceleration feature based on a new index type called a column store index. The new index type combined with new query operators processing batches of rows greatly improves data warehouse query performance: in some cases by hundreds of times and routinely a tenfold speedup for a broad range of decision support queries. Column store indexes are fully integrated with the rest of the system, including query processing and optimization. This paper gives an overview of the design and implementation of column store indexes including enhancements to query processing and query optimization to take full advantage of the new indexes. The resulting performance improvements are illustrated by a number of example queries. |
09a9a6b6a0b9e8fa210175587181d4a8329f3f20 | Learning, planning, and representing knowledge at multiple levels of temporal abstraction are key, longstanding challenges for AI. In this paper we consider how these challenges can be addressed within the mathematical framework of reinforcement learning and Markov decision processes (MDPs). We extend the usual notion of action in this framework to include options—closed-loop policies for taking action over a period of time. Examples of options include picking up an object, going to lunch, and traveling to a distant city, as well as primitive actions such as muscle twitches and joint torques. Overall, we show that options enable temporally abstract knowledge and action to be included in the reinforcement learning framework in a natural and general way. In particular, we show that options may be used interchangeably with primitive actions in planning methods such as dynamic programming and in learning methods such as Q-learning. Formally, a set of options defined over an MDP constitutes a semi-Markov decision process (SMDP), and the theory of SMDPs provides the foundation for the theory of options. However, the most interesting issues concern the interplay between the underlying MDP and the SMDP and are thus beyond SMDP theory. We present results for three such cases: (1) we show that the results of planning with options can be used during execution to interrupt options and thereby perform even better than planned, (2) we introduce new intra-option methods that are able to learn about an option from fragments of its execution, and (3) we propose a notion of subgoal that can be used to improve the options themselves. All of these results have precursors in the existing literature; the contribution of this paper is to establish them in a simpler and more general setting with fewer changes to the existing reinforcement learning framework. In particular, we show that these results can be obtained without committing to (or ruling out) any particular approach to state abstraction, hierarchy, function approximation, or the macroutility problem. ! 1999 Published by Elsevier Science B.V. All rights reserved. ∗ Corresponding author. 0004-3702/99/$ – see front matter ! 1999 Published by Elsevier Science B.V. All rights reserved. PII: S0004-3702(99)00052 -1 182 R.S. Sutton et al. / Artificial Intelligence 112 (1999) 181–211 |
3939607e665159002391082cd4a8952374d6bd99 | AbstructMultilevel voltage source converters are emerging as a new breed of power converter options for high-power applications. The multilevel voltage source converters typically synthesize the staircase voltage wave from several levels of dc capacitor voltages. One of the major limitations of the multilevel converters is the voltage unbalance between different levels. The techniques to balance the voltage between different levels normally involve voltage clamping or capacitor charge control. There are several ways of implementing voltage balance in multilevel converters. Without considering the traditional magnetic coupled converters, this paper presents three recently developed multilevel voltage source converters: 1) diode-clamp, 2) flyingcapacitors, and 3) cascaded-inverters with separate dc sources. The operating principle, features, constraints, and potential applications of these converters will be discussed. |
61c901789c7fb4721aae63769d7c4eb0e27a7e45 | Multilevel power converters that provide more than two levels of voltage to achieve smoother and less distorted ac-to-dc, dc-to-ac, and dc-to-dc power conversion, have attracted many contributors. This paper presents a generalized multilevel inverter (converter) topology with self voltage balancing. The existing multilevel inverters such as diode-clamped and capacitor-clamped multilevel inverters can be derived from the generalized inverter topology. Moreover, the generalized multilevel inverter topology provides a true multilevel structure that can balance each dc voltage level automatically without any assistance from other circuits, thus, in principle, providing a complete and true multilevel topology that embraces the existing multilevel inverters. From this generalized multilevel inverter topology, several new multilevel inverter structures can be derived. Some application examples of the generalized multilevel converter will be given. |
bb004d2d04ce6872d0f7965808ae4867aa037f8b | Electric power production in the 21 Century will see dramatic changes in both the physical infrastructure and the control and information infrastructure. A shift will take place from a relatively few large, concentrated generation centers and the transmission of electricity over mostly a high voltage ac grid (Fig. 1) to a more diverse and dispersed generation infrastructure that also has a higher percentage of dc transmission lines (Fig. 2) [1]. In the United States, generation capacity has not kept up with power demands, as over the last decade the reserve margins declined from 22% in 1990 to 16% in 1997. This declining margin trend is expected to continue over the next decade partly because of the uncertainty of how the future deregulated electrical environment will function. Less reserve margins will lead to less peaking capacity on highdemand days and more volatile energy prices [2]. This change in a physical infrastructure combined with a more deregulated electric power industry will result in more parties generating power – or distributed generation. Some of the distributed generation power sources that are expected to increase greatly their market share of the total power produced in the United States and abroad include renewable energy sources such as photovoltaics, wind, low-head hydro, and geothermal [3]. Fuel cell technology is also nearing the development point where it could start to supply a significant share of the power needs [4]. The advent of high power electronic modules has also encouraged the use of more dc transmission and made the prospects of interfacing dc power sources such as fuel cells and photovoltaics more easily attainable. A modular, scalable power electronics technology that is ideal for these types of utility applications is the transformerless multilevel converter [5]. |
c29ac3455cb73a777e1908b2c2c09b1d14ea1e9e | New residential scale photovoltaic (PV) arrays are commonly connected to the grid by a single dc-ac inverter connected to a series string of pv panels, or many small dc-ac inverters which connect one or two panels directly to the ac grid. This paper proposes an alternative topology of nonisolated per-panel dc-dc converters connected in series to create a high voltage string connected to a simplified dc-ac inverter. This offers the advantages of a "converter-per-panel" approach without the cost or efficiency penalties of individual dc-ac grid connected inverters. Buck, boost, buck-boost, and Cu/spl acute/k converters are considered as possible dc-dc converters that can be cascaded. Matlab simulations are used to compare the efficiency of each topology as well as evaluating the benefits of increasing cost and complexity. The buck and then boost converters are shown to be the most efficient topologies for a given cost, with the buck best suited for long strings and the boost for short strings. While flexible in voltage ranges, buck-boost, and Cu/spl acute/k converters are always at an efficiency or alternatively cost disadvantage. |
1bb06f401fac046234a9eeeb8735b8a456ba9d6d | In recent years, there has been a growing interest in wireless sensor networks. One of the major issues in wireless sensor network is developing an energy-efficient clustering protocol. Hierarchical clustering algorithms are very important in increasing the network’s life time. Each clustering algorithm is composed of two phases, the setup phase and steady state phase. The hot point in these algorithms is the cluster head selection. In this paper, we study the impact of heterogeneity of nodes in terms of their energy in wireless sensor networks that are hierarchically clustered. We assume that a percentage of the population of sensor nodes is equipped with the additional energy resources. We also assume that the sensor nodes are randomly distributed and are not mobile, the coordinates of the sink and the dimensions of the sensor field are known. Homogeneous clustering protocols assume that all the sensor nodes are equipped with the same amount of energy and as a result, they cannot take the advantage of the presence of node heterogeneity. Adapting this approach, we introduce an energy efficient heterogeneous clustered scheme for wireless sensor networks based on weighted election probabilities of each node to become a cluster head according to the residual energy in each node. Finally, the simulation results demonstrate that our proposed heterogeneous clustering approach is more effective in prolonging the network lifetime compared with LEACH. 2008 Elsevier B.V. All rights reserved. |
0352893287ea6c7d6a65946706b1b75cbe598798 | |
552528ae817834765e491d8f783c785c07ca7ccb | Privacy of social network data is a growing concern which threatens to limit access to this valuable data source. Analysis of the graph structure of social networks can provide valuable information for revenue generation and social science research, but unfortunately, ensuring this analysis does not violate individual privacy is difficult. Simply anonymizing graphs or even releasing only aggregate results of analysis may not provide sufficient protection. Differential privacy is an alternative privacy model, popular in data-mining over tabular data, which uses noise to obscure individuals' contributions to aggregate results and offers a very strong mathematical guarantee that individuals' presence in the data-set is hidden. Analyses that were previously vulnerable to identification of individuals and extraction of private data may be safely released under differential-privacy guarantees. We review two existing standards for adapting differential privacy to network data and analyse the feasibility of several common social-network analysis techniques under these standards. Additionally, we propose out-link privacy, a novel standard for differential privacy over network data, and introduce two powerful out-link private algorithms for common network analysis techniques that were infeasible to privatize under previous differential privacy standards. |
002c3339df17101b1b8f56d534ba4de2437f7a22 | We perform fast vehicle detection from traffic surveillance cameras. A novel deep learning framework, namely Evolving Boxes, is developed that proposes and refines the object boxes under different feature representations. Specifically, our framework is embedded with a light-weight proposal network to generate initial anchor boxes as well as to early discard unlikely regions; a fine-turning network produces detailed features for these candidate boxes. We show intriguingly that by applying different feature fusion techniques, the initial boxes can be refined for both localization and recognition. We evaluate our network on the recent DETRAC benchmark and obtain a significant improvement over the state-of-the-art Faster RCNN by 9.5% mAP. Further, our network achieves 9–13 FPS detection speed on a moderate commercial GPU. |
541a89579d77942c0b8ac6debf8ef99a04684dca | In this paper, the design of a series-fed microstrip antenna array is discussed and its design challenge of limited gain is explained. It discusses that the phase between the patches of the array plays a significant role in overcoming the challenge. A 23 element series fed linear array at Ka-band gives gain of 19dBi with beam in broadside direction and has SLL better than -15dB. A scaled down version at C-band has been designed and fabricated. A 7-element series fed antenna array yielded measured gain of 15.1 dBi at 5.79 GHz. This design can be extended for planar antenna arrays to obtain higher gain. |
056e13d5045e7d594489705f78834cfaf6642c36 | Boosting is a general method for improving the accuracy of any given learning algorithm. Focusing primarily on the AdaBoost algorithm, this chapter overviews some of the recent work on boosting including analyses of AdaBoost’s training error and generalization error; boosting’s connection to game theory and linear programming; the relationship between boosting and logistic regression; extensions of AdaBoost for multiclass classification problems; methods of incorporating human knowledge into boosting; and experimental and applied work using boosting. |
38f35dd624cd1cf827416e31ac5e0e0454028eca | We introduce DropConnect, a generalization of Dropout (Hinton et al., 2012), for regularizing large fully-connected layers within neural networks. When training with Dropout, a randomly selected subset of activations are set to zero within each layer. DropConnect instead sets a randomly selected subset of weights within the network to zero. Each unit thus receives input from a random subset of units in the previous layer. We derive a bound on the generalization performance of both Dropout and DropConnect. We then evaluate DropConnect on a range of datasets, comparing to Dropout, and show state-of-the-art results on several image recognition benchmarks by aggregating multiple DropConnect-trained models. |
8579b32d0cabee5a9f41dcde0b17aa51d9508850 | We have developed a gesture recognition system, based on Hidden Markov Models, which can interactively recognize gestures and perform online learning of new gestures. In addition, it is able to update its model of a gesture iteratively with each example it recognizes. This system has demonstrated reliable recognition of 14 di erent gestures after only one or two examples of each. The system is currently interfaced to a Cyberglove for use in recognition of gestures from the sign language alphabet. The system is being implemented as part of an interactive interface for robot teleoperation and programming by example. |
0617301c077e56c44933e2b790a270f3e590db12 | We present a new indexing method for the approximate string matching problem. The method is based on a suffix array combined with a partitioning of the pattern. We analyze the resulting algorithm and show that the average retrieval time is , for some that depends on the error fraction tolerated and the alphabet size . It is shown that for approximately , where . The space required is four times the text size, which is quite moderate for this problem. We experimentally show that this index can outperform by far all the existing alternatives for indexed approximate searching. These are also the first experiments that compare the different existing schemes. |
6c94ec89603ffafcdba4600bc7506a2df35cc246 | This Doctoral Consortium paper discusses how NLP can be applied in the domain of BPM in order to automatically generate business process models from existing documentation within the organization. The main idea is that from the syntactic and grammatical structure of a sentence, the components of a business process model can be derived (i.e. activities, resources, tasks, patterns). The result would be a business process model depicted using BPMN - a dedicated business process modeling technique |
075bc988728788aa033b04dee1753ded711180ee | We consider the problem of automatically recognizing human faces from frontal views with varying expression and illumination, as well as occlusion and disguise. We cast the recognition problem as one of classifying among multiple linear regression models and argue that new theory from sparse signal representation offers the key to addressing this problem. Based on a sparse representation computed by C1-minimization, we propose a general classification algorithm for (image-based) object recognition. This new framework provides new insights into two crucial issues in face recognition: feature extraction and robustness to occlusion. For feature extraction, we show that if sparsity in the recognition problem is properly harnessed, the choice of features is no longer critical. What is critical, however, is whether the number of features is sufficiently large and whether the sparse representation is correctly computed. Unconventional features such as downsampled images and random projections perform just as well as conventional features such as eigenfaces and Laplacianfaces, as long as the dimension of the feature space surpasses certain threshold, predicted by the theory of sparse representation. This framework can handle errors due to occlusion and corruption uniformly by exploiting the fact that these errors are often sparse with respect to the standard (pixel) basis. The theory of sparse representation helps predict how much occlusion the recognition algorithm can handle and how to choose the training images to maximize robustness to occlusion. We conduct extensive experiments on publicly available databases to verify the efficacy of the proposed algorithm and corroborate the above claims. |
0d117c9fc3393237d71be5dce9bb6498b5e0c020 | Given i.i.d. observations of a random vector X ∈ R, we study the problem of estimating both its covariance matrix Σ, and its inverse covariance or concentration matrix Θ = (Σ). When X is multivariate Gaussian, the non-zero structure of Θ is specified by the graph of an associated Gaussian Markov random field; and a popular estimator for such sparse Θ is the l1-regularized Gaussian MLE. This estimator is sensible even for for non-Gaussian X, since it corresponds to minimizing an l1-penalized log-determinant Bregman divergence. We analyze its performance under high-dimensional scaling, in which the number of nodes in the graph p, the number of edges s, and the maximum node degree d, are allowed to grow as a function of the sample size n. In addition to the parameters (p, s, d), our analysis identifies other key quantities that control rates: (a) the l∞-operator norm of the true covariance matrix Σ; and (b) the l∞ operator norm of the sub-matrix Γ∗SS , where S indexes the graph edges, and Γ∗ = (Θ) ⊗ (Θ); and (c) a mutual incoherence or irrepresentability measure on the matrix Γ∗ and (d) the rate of decay 1/f(n, δ) on the probabilities {|Σ̂ij − Σ∗ij | > δ}, where Σ̂ is the sample covariance based on n samples. Our first result establishes consistency of our estimate Θ̂ in the elementwise maximum-norm. This in turn allows us to derive convergence rates in Frobenius and spectral norms, with improvements upon existing results for graphs with maximum node degrees d = o( √ s). In our second result, we show that with probability converging to one, the estimate Θ̂ correctly specifies the zero pattern of the concentration matrix Θ∗. We illustrate our theoretical results via simulations for various graphs and problem parameters, showing good correspondences between the theoretical predictions and behavior in simulations. AMS 2000 subject classifications: Primary 62F12; secondary 62F30. |
216c8515f9f53533b2e87c8183e70d3b50c2c097 | A popular approach in current commercial anti-malware software detects malicious programs by searching in the code of programs for scan strings that are byte sequences indicative of malicious code. The scan strings, also known as the signatures of existing malware, are extracted by malware analysts from known malware samples, and stored in a database often referred to as a virus dictionary. This process often involves a significant amount of human efforts. In addition, there are two major limitations in this technique. First, not all malicious programs have bit patterns that are evidence of their malicious nature. Therefore, some malware is not recorded in the virus dictionary and can not be detected through signature matching. Second, searching for specific bit patterns will not work on malware that can take many forms--obfuscated malware. Signature matching has been shown to be incapable of identifying new malware patterns and fails to recognize obfuscated malware. This paper presents a malware detection technique that discovers malware by means of a learning engine trained on a set of malware instances and a set of benign code instances. The learning engine uses an adaptive data compression model--prediction by partial matching (PPM)--to build two compression models, one from the malware instances and the other from the benign code instances. A code instance is classified, either as "malware" or "benign", by minimizing its estimated cross entropy. Our preliminary results are very promising. We achieved about 0.94 true positive rate with as low as 0.016 false positive rate. Our experiments also demonstrate that this technique can effectively detect unknown and obfuscated malware. |
682640754c867ecc6ae2ccaa5dc68403ee7d2e63 | We show how to realize two-factor authentication for a Bitcoin wallet. To do so, we explain how to employ an ECDSA adaption of the two-party signature protocol by MacKenzie and Reiter (Int J Inf Secur 2(3–4):218–239, 2004. doi: 10.1007/s10207-004-0041-0 ) in the context of Bitcoin and present a prototypic implementation of a Bitcoin wallet that offers both: two-factor authentication and verification over a separate channel. Since we use a smart phone as the second authentication factor, our solution can be used with hardware already available to most users and the user experience is quite similar to the existing online banking authentication methods. |
e427c8d3c1b616d319c8b5f233e725d4ebfd9768 | Localizing functional regions of objects or affordances is an important aspect of scene understanding and relevant for many robotics applications. In this work, we introduce a pixel-wise annotated affordance dataset of 3090 images containing 9916 object instances. Since parts of an object can have multiple affordances, we address this by a convolutional neural network for multilabel affordance segmentation. We also propose an approach to train the network from very few keypoint annotations. Our approach achieves a higher affordance detection accuracy than other weakly supervised methods that also rely on keypoint annotations or image annotations as weak supervision. |
731359cd04c7625549e78d9fbd9ad362642be9c8 | Uncertainty, its modelling and analysis have been discussed across many literatures including statistics and operational research, knowledge management and philosophy. Adherents to Bayesian approaches have usually argued that uncertainty should either be modelled by probabilities or resolved by discussion which clarifies meaning. Others have followed Knight in distinguishing between contexts of risk and of uncertainty: the former admitting modelling and analysis through probability; the latter not. There are also host of approaches in the literatures stemming from Zadeh’s concept of a fuzzy set. Theories of sense-making in the philosophy and management literatures see knowledge and uncertainty as opposite extremes of human understanding and discuss the resolution of uncertainty accordingly. Here we take a Bayesian stance, but a softer one than the conventional, which recognises the concerns in other approaches and, in particular, sets their concerns in the Cynefin framework of decision contexts to reflect on processes of modelling and analysis in statistical, risk and decision analysis. The approach builds on several recent strands of discussion which argue for a convergence of qualitative scenario planning ideas and more quantitative approaches to analysis I discuss how these suggestions and discussions relate to some earlier thinking on the methodology of modelling and, in particular, the concept of a ‘small world’ articulated by Savage. |
f0fe2b9cb1f4de756db573127fe7560421e5de3d | For over two decades, cannabis, commonly known as marijuana, has been the most widely used illicit drug by young people in high-income countries, and has recently become popular on a global scale. Epidemiological research during the past 10 years suggests that regular use of cannabis during adolescence and into adulthood can have adverse effects. Epidemiological, clinical, and laboratory studies have established an association between cannabis use and adverse outcomes. We focus on adverse health effects of greatest potential public health interest-that is, those that are most likely to occur and to affect a large number of cannabis users. The most probable adverse effects include a dependence syndrome, increased risk of motor vehicle crashes, impaired respiratory function, cardiovascular disease, and adverse effects of regular use on adolescent psychosocial development and mental health. |
4d625677469be99e0a765a750f88cfb85c522cce | Our goal is to automate the understanding of natural hand-object manipulation by developing computer visionbased techniques. Our hypothesis is that it is necessary to model the grasp types of hands and the attributes of manipulated objects in order to accurately recognize manipulation actions. Specifically, we focus on recognizing hand grasp types, object attributes and actions from a single image within an unified model. First, we explore the contextual relationship between grasp types and object attributes, and show how that context can be used to boost the recognition of both grasp types and object attributes. Second, we propose to model actions with grasp types and object attributes based on the hypothesis that grasp types and object attributes contain complementary information for characterizing different actions. Our proposed action model outperforms traditional appearance-based models which are not designed to take into account semantic constraints such as grasp types or object attributes. Experiment results on public egocentric activities datasets strongly support our hypothesis. |
59fa9d40d129f18f1f3193f65935fcb9e2042afb | For new automotive applications and services, information technology (IT) has gained central importance. IT-related costs in car manufacturing are already high and they will increase dramatically in the future. Yet whereas safety and reliability have become a relatively well-established field, the protection of vehicular IT systems against systematic manipulation or intrusion has only recently started to emerge. Nevertheless, IT security is already the base of some vehicular applications such as immobilizers or digital tachographs. To securely enable future automotive applications and business models, IT security will be one of the central technologies for the next generation of vehicles. After a state-of-the-art overview of IT security in vehicles, we give a short introduction into cryptographic terminology and functionality. This contribution will then identify the need for automotive IT security while presenting typical attacks, resulting security objectives, and characteristic constraints within the automotive area. We will introduce core security technologies and relevant security mechanisms followed by a detailed description of critical vehicular applications, business models, and components relying on IT security. We conclude our contribution with a detailed statement about challenges and opportunities for the automotive IT community for embedding IT security in vehicles. |
ae5e5085b4e8f4851d9fd76e1d3845da942c3147 | When you drive to somewhere far away, you will leave your current location via one of only a few important traffic junctions. Starting from this informal observation, we developed an algorithmic approach, transit node routing, that allows us to reduce quickest path queries in road networks to a small number of table lookups. For road maps of Western Europe and the United States, our best query times improved over the best previously published figures by two orders of magnitude. This is also more than one million times faster than the best known algorithm for general networks. |
855d0f722d75cc56a66a00ede18ace96bafee6bd | Theano is a linear algebra compiler that optimizes a user’s symbolically-specified mathematical computations to produce efficient low-level implementations. In this paper, we present new features and efficiency improvements to Theano, and benchmarks demonstrating Theano’s performance relative to Torch7, a recently introduced machine learning library, and to RNNLM, a C++ library targeted at recurrent neural networks. |
f97b592092377a6e3afeb2f55e2aacd8794cb8b2 | A compact dual-band microstrip bandpass filter is proposed and designed to operate at 2.4 and 5.2GHz without needing any external impedance-matching block. The modified half-wavelength stepped-impedance resonator with sinuous configuration is constructed to simultaneously excite the dual resonances at these two specified frequencies with miniaturized overall size. The parallel-coupled microstrip line is properly characterized to minimize the return losses within both dual passbands. The optimized results exhibit the good dual-band filtering performances with return losses higher than 20dB as well being confirmed by experimentation with a fabricated filter circuit. |
d4a8e93f004c86267eead89edecbd332518dbf21 | SDM is a high-level semantics-based database description and structuring formalism (database model) for databases. This database model is designed to capture more of the meaning of an application environment than is possible with contemporary database models. An SDM specification describes a database in terms of the kinds of entities that exist in the application environment, the classifications and groupings of those entities, and the structural interconnections among them. SDM provides a collection of high-level modeling primitives to capture the semantics of an application environment. By accommodating derived information in a database structural specification, SDM allows the same information to be viewed in several ways; this makes it possible to directly accommodate the variety of needs and processing requirements typically present in database applications. The design of the present SDM is based on our experience in using a preliminary version of it.
SDM is designed to enhance the effectiveness and usability of database systems. An SDM database description can serve as a formal specification and documentation tool for a database; it can provide a basis for supporting a variety of powerful user interface facilities, it can serve as a conceptual database model in the database design process; and, it can be used as the database model for a new kind of database management system. |
cced93571a42beb3853c2d0b3217b3f57b614bb0 | |
eefcc7bcc05436dac9881acb4ff4e4a0b730e175 | We address image classification on a large-scale, i.e. when a large number of images and classes are involved. First, we study classification accuracy as a function of the image signature dimensionality and the training set size. We show experimentally that the larger the training set, the higher the impact of the dimensionality on the accuracy. In other words, high-dimensional signatures are important to obtain state-of-the-art results on large datasets. Second, we tackle the problem of data compression on very large signatures (on the order of 105 dimensions) using two lossy compression strategies: a dimensionality reduction technique known as the hash kernel and an encoding technique based on product quantizers. We explain how the gain in storage can be traded against a loss in accuracy and/or an increase in CPU cost. We report results on two large databases — ImageNet and a dataset of lM Flickr images — showing that we can reduce the storage of our signatures by a factor 64 to 128 with little loss in accuracy. Integrating the decompression in the classifier learning yields an efficient and scalable training algorithm. On ILSVRC2010 we report a 74.3% accuracy at top-5, which corresponds to a 2.5% absolute improvement with respect to the state-of-the-art. On a subset of 10K classes of ImageNet we report a top-1 accuracy of 16.7%, a relative improvement of 160% with respect to the state-of-the-art. |
016335ce7e0a073623e1deac7138b28913dbf594 | People learning new concepts can often generalize successfully from just a single example, yet machine learning algorithms typically require tens or hundreds of examples to perform with similar accuracy. People can also use learned concepts in richer ways than conventional algorithms—for action, imagination, and explanation. We present a computational model that captures these human learning abilities for a large class of simple visual concepts: handwritten characters from the world’s alphabets. The model represents concepts as simple programs that best explain observed examples under a Bayesian criterion. On a challenging one-shot classification task, the model achieves human-level performance while outperforming recent deep learning approaches. We also present several “visual Turing tests” probing the model’s creative generalization abilities, which in many cases are indistinguishable from human behavior. |
83bcd3591d8e5d43d65e9e1e83e4c257f8431d4a | A simple technique is developed in this communication to increase the axial ratio (AR) bandwidth and achieve good impedance matching of a single-fed low profile circularly polarized (CP) stacked patch antenna. The proposed antenna is composed of a driven patch layer and a parasitic patch layer. The driven patch layer consists of a truncated main patch, a parasitic patch and a probe feeding structure while the stacked patch layer is comprised of five patches. The proposed antenna combines the attractive features such as low profile, wide impedance and AR bandwidths, high gain as well as easiness of design, manufacture and integration. The antenna operating at 6 GHz band is designed and fabricated on an FR4 substrate and the overall volume is 0.8 λ0 ×0.8 λ0 ×0.09 λ0. Measured results show that the antenna achieves an impedance bandwidth of more than 30% for dB, a 3-dB AR bandwidth of about 20.7%, and a gain level of over 7.9 dBi within the 3-dB AR bandwidth. |
b8deb0fcfa16a0dcb46652240c015cef93c711ed | Crohn's disease may involve all parts of the gastrointestinal tract and may often involve other organs as well. These non-intestinal affections are termed extraintestinal manifestations. Vulval involvement is an uncommon extraintestinal manifestation of Crohn's disease, and it is very rare in children. Patients with vulval CD typically present with erythema and edema of the labia majora, which progresses to extensive ulcer formation. Vulval Crohn's disease can appear before or after intestinal problems or it may occur simultaneously. We present a 10-year-old girl with intestinal Crohn's disease complicated with perianal skin tags and asymptomatic unilateral labial hypertrophy. The course of her lesion was independent of the intestinal disease and responded significantly to medical treatment including azathioprine and topical steroid. We emphasize that although vulval involvement in childhood is uncommon, Crohn's disease must be considered in the differential diagnosis of nontender, red, edematous lesions of the genital area. |
9f908c018a0c57bab2a4a3ee723c818e893fcb1b | |
2fdee22266d58ae4e891711208106ca46c8e2778 | Product quantization is an effective vector quantization approach to compactly encode high-dimensional vectors for fast approximate nearest neighbor (ANN) search. The essence of product quantization is to decompose the original high-dimensional space into the Cartesian product of a finite number of low-dimensional subspaces that are then quantized separately. Optimal space decomposition is important for the performance of ANN search, but still remains unaddressed. In this paper, we optimize product quantization by minimizing quantization distortions w.r.t. the space decomposition and the quantization codebooks. We present two novel methods for optimization: a non-parametric method that alternatively solves two smaller sub-problems, and a parametric method that is guaranteed to achieve the optimal solution if the input data follows some Gaussian distribution. We show by experiments that our optimized approach substantially improves the accuracy of product quantization for ANN search. |
149d8514b026cca3b31ef8379e78aeb7c7795eb7 | With the adoption of 2G and 3G cellular network technologies, mobile phones now have the bandwidth capability to stream data back to monitoring stations in real-time. Our paper describes the design and evaluation of a Bluetooth electrocardiogram sensor that transmits medical data to a cell phone. This data is displayed and stored on the phone. Future development of the system will relay this data over a cellular GPRS network. The current system provides a low cost and lightweight alternative to existing EKG event monitors. The final GPRS connected system will provide continuous monitoring of a patient 's heart anywhere cellular coverage is available |
226ceb666cdb2090fc3ab786129e83f3ced56e05 | Neural machine translation NMT has been prominent in many machine translation tasks. However, in some domain-specific tasks, only the corpora from similar domains can improve translation performance. If out-of-domain corpora are directly added into the in-domain corpus, the translation performance may even degrade. Therefore, domain adaptation techniques are essential to solve the NMT domain problem. Most existing methods for domain adaptation are designed for the conventional phrase-based machine translation. For NMT domain adaptation, there have been only a few studies on topics such as fine tuning, domain tags, and domain features. In this paper, we have four goals for sentence level NMT domain adaptation. First, the NMT's internal sentence embedding is exploited and the sentence embedding similarity is used to select out-of-domain sentences that are close to the in-domain corpus. Second, we propose three sentence weighting methods, i.e., sentence weighting, domain weighting, and batch weighting, to balance the data distribution during NMT training. Third, in addition, we propose dynamic training methods to adjust the sentence selection and weighting during NMT training. Fourth, to solve the multidomain problem in a real-world NMT scenario where the domain distributions of training and testing data often mismatch, we proposed a multidomain sentence weighting method to balance the domain distributions of training data and match the domain distributions of training and testing data. The proposed methods are evaluated in international workshop on spoken language translation IWSLT English-to-French/German tasks and a multidomain English-to-French task. Empirical results show that the sentence selection and weighting methods can significantly improve the NMT performance, outperforming the existing baselines. |
Subsets and Splits