Unnamed: 0.1
int64
0
41k
Unnamed: 0
int64
0
41k
author
stringlengths
9
1.39k
id
stringlengths
11
18
summary
stringlengths
25
3.66k
title
stringlengths
4
258
year
int64
1.99k
2.02k
arxiv_url
stringlengths
32
39
info
stringlengths
523
3.18k
embeddings
stringlengths
16.9k
17.1k
0
0
['Ahmed Osman', 'Wojciech Samek']
1802.00209v1
We propose an architecture for VQA which utilizes recurrent layers to generate visual and textual attention. The memory characteristic of the proposed recurrent attention units offers a rich joint embedding of visual and textual features and enables the model to reason relations between several parts of the image and q...
Dual Recurrent Attention Units for Visual Question Answering
2,018
http://arxiv.org/pdf/1802.00209v1
Title Dual Recurrent Attention Units Visual Question Answering Summary propose architecture VQA utilizes recurrent layer generate visual textual attention memory characteristic proposed recurrent attention unit offer rich joint embedding visual textual feature enables model reason relation several part image question s...
[0.030238820239901543, 0.016444332897663116, -0.016811851412057877, 0.06810687482357025, 0.0009400892304256558, 0.01178388949483633, -0.0018757805228233337, -0.009158705361187458, 0.002846865216270089, -0.038819942623376846, 0.0050158933736383915, -0.03350161761045456, -0.021443059667944908, 0.050835151225328445, 0.053...
1
1
['Ji Young Lee', 'Franck Dernoncourt']
1603.03827v1
Recent approaches based on artificial neural networks (ANNs) have shown promising results for short-text classification. However, many short texts occur in sequences (e.g., sentences in a document or utterances in a dialog), and most existing ANN-based systems do not leverage the preceding short texts when classifying ...
Sequential Short-Text Classification with Recurrent and Convolutional Neural Networks
2,016
http://arxiv.org/pdf/1603.03827v1
Title Sequential ShortText Classification Recurrent Convolutional Neural Networks Summary Recent approach based artificial neural network ANNs shown promising result shorttext classification However many short text occur sequence eg sentence document utterance dialog existing ANNbased system leverage preceding short te...
[0.040623173117637634, 0.010163335129618645, 0.0038399931509047747, 0.06766032427549362, -0.01700439304113388, -0.003362833522260189, 0.025043299421668053, 0.020645055919885635, 0.03404254466295242, -0.081425741314888, -0.029075222089886665, -0.04023962467908859, 0.01441923063248396, 0.06081303581595421, 0.004792141262...
2
2
['Iulian Vlad Serban', 'Tim Klinger', 'Gerald Tesauro', 'Kartik Talamadupula', 'Bowen Zhou', 'Yoshua Bengio', 'Aaron Courville']
1606.00776v2
We introduce the multiresolution recurrent neural network, which extends the sequence-to-sequence framework to model natural language generation as two parallel discrete stochastic processes: a sequence of high-level coarse tokens, and a sequence of natural language tokens. There are many ways to estimate or learn the ...
Multiresolution Recurrent Neural Networks: An Application to Dialogue Response Generation
2,016
http://arxiv.org/pdf/1606.00776v2
Title Multiresolution Recurrent Neural Networks Application Dialogue Response Generation Summary introduce multiresolution recurrent neural network extends sequencetosequence framework model natural language generation two parallel discrete stochastic process sequence highlevel coarse token sequence natural language to...
[0.07278219610452652, 0.03251509740948677, -0.0008744823280721903, 0.02868502028286457, -0.0180113073438406, -0.007289289031177759, -0.0035347219090908766, -0.01468235719949007, -0.006232827436178923, -0.07925242930650711, 0.02225460298359394, -0.025085624307394028, 0.0075123305432498455, 0.07406775653362274, -0.001531...
3
3
['Sebastian Ruder', 'Joachim Bingel', 'Isabelle Augenstein', 'Anders Søgaard']
1705.08142v2
Multi-task learning is motivated by the observation that humans bring to bear what they know about related problems when solving new ones. Similarly, deep neural networks can profit from related tasks by sharing parameters with other networks. However, humans do not consciously decide to transfer knowledge between task...
Learning what to share between loosely related tasks
2,017
http://arxiv.org/pdf/1705.08142v2
Title Learning share loosely related task Summary Multitask learning motivated observation human bring bear know related problem solving new one Similarly deep neural network profit related task sharing parameter network However human consciously decide transfer knowledge task Natural Language Processing NLP hard predi...
[0.022487860172986984, 0.03934193029999733, -0.032645177096128464, 0.00894354097545147, -0.02416212111711502, -0.02859516069293022, 0.05892830342054367, -0.02223842963576317, -0.02291261963546276, -0.0076596797443926334, -0.08599571883678436, 0.01687566377222538, -0.04040209576487541, 0.07899221777915955, 0.02695311792...
4
4
['Iulian V. Serban', 'Chinnadhurai Sankar', 'Mathieu Germain', 'Saizheng Zhang', 'Zhouhan Lin', 'Sandeep Subramanian', 'Taesup Kim', 'Michael Pieper', 'Sarath Chandar', 'Nan Rosemary Ke', 'Sai Rajeshwar', 'Alexandre de Brebisson', 'Jose M. R. Sotelo', 'Dendi Suhubdy', 'Vincent Michalski', 'Alexandre Nguyen', 'Joelle Pi...
1709.02349v2
We present MILABOT: a deep reinforcement learning chatbot developed by the Montreal Institute for Learning Algorithms (MILA) for the Amazon Alexa Prize competition. MILABOT is capable of conversing with humans on popular small talk topics through both speech and text. The system consists of an ensemble of natural langu...
A Deep Reinforcement Learning Chatbot
2,017
http://arxiv.org/pdf/1709.02349v2
Title Deep Reinforcement Learning Chatbot Summary present MILABOT deep reinforcement learning chatbot developed Montreal Institute Learning Algorithms MILA Amazon Alexa Prize competition MILABOT capable conversing human popular small talk topic speech text system consists ensemble natural language generation retrieval ...
[0.08369601517915726, 0.020538926124572754, -0.006166993174701929, -0.02317694202065468, -0.016374515369534492, 0.007429464254528284, -0.004607527516782284, 0.004775070585310459, 0.014056166633963585, -0.020789040252566338, -0.023264795541763306, -0.004501406103372574, -0.025124873965978622, 0.09215951710939407, -0.004...
5
5
['Kelvin Guu', 'Tatsunori B. Hashimoto', 'Yonatan Oren', 'Percy Liang']
1709.08878v1
We propose a new generative model of sentences that first samples a prototype sentence from the training corpus and then edits it into a new sentence. Compared to traditional models that generate from scratch either left-to-right or by first sampling a latent sentence vector, our prototype-then-edit model improves perp...
Generating Sentences by Editing Prototypes
2,017
http://arxiv.org/pdf/1709.08878v1
Title Generating Sentences Editing Prototypes Summary propose new generative model sentence first sample prototype sentence training corpus edits new sentence Compared traditional model generate scratch either lefttoright first sampling latent sentence vector prototypethenedit model improves perplexity language modelin...
[0.08618932217359543, 0.04899665713310242, -0.02246158942580223, 0.019272757694125175, -0.03703729435801506, 0.01352265290915966, 0.037966545671224594, -0.024638528004288673, -0.027826817706227303, -0.03432176634669304, 0.004921475891023874, -0.011469664983451366, -0.023666782304644585, 0.02391095645725727, 0.040919352...
6
6
['Iulian V. Serban', 'Chinnadhurai Sankar', 'Mathieu Germain', 'Saizheng Zhang', 'Zhouhan Lin', 'Sandeep Subramanian', 'Taesup Kim', 'Michael Pieper', 'Sarath Chandar', 'Nan Rosemary Ke', 'Sai Rajeswar', 'Alexandre de Brebisson', 'Jose M. R. Sotelo', 'Dendi Suhubdy', 'Vincent Michalski', 'Alexandre Nguyen', 'Joelle Pin...
1801.06700v1
We present MILABOT: a deep reinforcement learning chatbot developed by the Montreal Institute for Learning Algorithms (MILA) for the Amazon Alexa Prize competition. MILABOT is capable of conversing with humans on popular small talk topics through both speech and text. The system consists of an ensemble of natural langu...
A Deep Reinforcement Learning Chatbot (Short Version)
2,018
http://arxiv.org/pdf/1801.06700v1
Title Deep Reinforcement Learning Chatbot Short Version Summary present MILABOT deep reinforcement learning chatbot developed Montreal Institute Learning Algorithms MILA Amazon Alexa Prize competition MILABOT capable conversing human popular small talk topic speech text system consists ensemble natural language generat...
[0.07682596892118454, 0.024237127974629402, -0.0078873410820961, -0.021940061822533607, -0.00957291666418314, 0.009843532927334309, 0.0012712820898741484, -0.0006560627953149378, 0.006107945926487446, -0.02412767894566059, -0.021866416558623314, -0.002529066288843751, -0.02169090323150158, 0.08880618214607239, -0.00294...
7
7
['Darko Brodic', 'Alessia Amelio', 'Zoran N. Milivojevic', 'Milena Jevtic']
1609.06492v1
The paper introduces a new method for discrimination of documents given in different scripts. The document is mapped into a uniformly coded text of numerical values. It is derived from the position of the letters in the text line, based on their typographical characteristics. Each code is considered as a gray level. Ac...
Document Image Coding and Clustering for Script Discrimination
2,016
http://arxiv.org/pdf/1609.06492v1
Title Document Image Coding Clustering Script Discrimination Summary paper introduces new method discrimination document given different script document mapped uniformly coded text numerical value derived position letter text line based typographical characteristic code considered gray level Accordingly coded text dete...
[0.015126252546906471, 0.0003478110593277961, -0.015845399349927902, 0.04707137867808342, -0.02666432224214077, 0.02335183322429657, 0.03511674329638481, 0.1077132597565651, 0.03802177309989929, -0.0408138744533062, 0.0360984206199646, 0.020254211500287056, 0.042206279933452606, 0.03312941640615463, -0.0029164124280214...
8
8
['Mateusz Malinowski', 'Mario Fritz']
1610.01076v1
Together with the development of more accurate methods in Computer Vision and Natural Language Understanding, holistic architectures that answer on questions about the content of real-world images have emerged. In this tutorial, we build a neural-based approach to answer questions about images. We base our tutorial on ...
Tutorial on Answering Questions about Images with Deep Learning
2,016
http://arxiv.org/pdf/1610.01076v1
Title Tutorial Answering Questions Images Deep Learning Summary Together development accurate method Computer Vision Natural Language Understanding holistic architecture answer question content realworld image emerged tutorial build neuralbased approach answer question image base tutorial two datasets mostly DAQUAR bit...
[0.05145927891135216, 0.03849175572395325, -0.02072332054376602, 0.07118497788906097, -0.004714971873909235, -0.005271739326417446, 0.017641786485910416, -0.0026514835190027952, -0.03388672694563866, -0.031107222661376, -0.0179398525506258, -0.010696107521653175, 0.008540982380509377, 0.07507918775081635, 0.00251430040...
9
9
['Tony Beltramelli']
1705.07962v2
Transforming a graphical user interface screenshot created by a designer into computer code is a typical task conducted by a developer in order to build customized software, websites, and mobile applications. In this paper, we show that deep learning methods can be leveraged to train a model end-to-end to automatically...
pix2code: Generating Code from a Graphical User Interface Screenshot
2,017
http://arxiv.org/pdf/1705.07962v2
Title pix2code Generating Code Graphical User Interface Screenshot Summary Transforming graphical user interface screenshot created designer computer code typical task conducted developer order build customized software website mobile application paper show deep learning method leveraged train model endtoend automatica...
[0.020227601751685143, 0.03282645717263222, -0.03515835851430893, 0.026191987097263336, -0.02822529338300228, -0.009803039021790028, 0.056261561810970306, 0.0431673526763916, -0.04715876281261444, -0.03450101986527443, -0.0003499208833090961, 0.03791843354701996, 0.019206956028938293, 0.16242928802967072, 0.01970106549...
10
10
['Fred Richardson', 'Douglas Reynolds', 'Najim Dehak']
1504.00923v1
Learned feature representations and sub-phoneme posteriors from Deep Neural Networks (DNNs) have been used separately to produce significant performance gains for speaker and language recognition tasks. In this work we show how these gains are possible using a single DNN for both speaker and language recognition. The u...
A Unified Deep Neural Network for Speaker and Language Recognition
2,015
http://arxiv.org/pdf/1504.00923v1
Title Unified Deep Neural Network Speaker Language Recognition Summary Learned feature representation subphoneme posterior Deep Neural Networks DNNs used separately produce significant performance gain speaker language recognition task work show gain possible using single DNN speaker language recognition unified DNN ap...
[0.0012832034844905138, 0.06519152224063873, 0.016510702669620514, 0.05018047243356705, 0.003570161061361432, 0.020978888496756554, 0.06571382284164429, -0.016524504870176315, -0.008365627378225327, 0.006050108931958675, -0.06710567325353622, -0.05633586272597313, 0.04013851284980774, 0.002399613382294774, -0.019586972...
11
11
['Hieu Pham', 'Melody Y. Guan', 'Barret Zoph', 'Quoc V. Le', 'Jeff Dean']
1802.03268v2
We propose Efficient Neural Architecture Search (ENAS), a fast and inexpensive approach for automatic model design. In ENAS, a controller learns to discover neural network architectures by searching for an optimal subgraph within a large computational graph. The controller is trained with policy gradient to select a su...
Efficient Neural Architecture Search via Parameter Sharing
2,018
http://arxiv.org/pdf/1802.03268v2
Title Efficient Neural Architecture Search via Parameter Sharing Summary propose Efficient Neural Architecture Search ENAS fast inexpensive approach automatic model design ENAS controller learns discover neural network architecture searching optimal subgraph within large computational graph controller trained policy gr...
[0.002730378182604909, 0.04860375449061394, -0.025265756994485855, 0.05351081117987633, 0.013431431725621223, -0.020422853529453278, 0.009471342898905277, 0.004952882416546345, 0.010119565762579441, 0.01924092136323452, -0.05238935723900795, 0.031350985169410706, -0.023592302575707436, 0.04062184318900108, 0.0388815924...
12
12
['Brenden M. Lake', 'Tomer D. Ullman', 'Joshua B. Tenenbaum', 'Samuel J. Gershman']
1604.00289v3
Recent progress in artificial intelligence (AI) has renewed interest in building systems that learn and think like people. Many advances have come from using deep neural networks trained end-to-end in tasks such as object recognition, video games, and board games, achieving performance that equals or even beats humans ...
Building Machines That Learn and Think Like People
2,016
http://arxiv.org/pdf/1604.00289v3
Title Building Machines Learn Think Like People Summary Recent progress artificial intelligence AI renewed interest building system learn think like people Many advance come using deep neural network trained endtoend task object recognition video game board game achieving performance equal even beat human respect Despi...
[0.031925417482852936, 0.050153762102127075, -0.04506157338619232, 0.025901375338435173, 0.010624541901051998, 0.0023722907062619925, 0.02059454284608364, -0.024825099855661392, 0.002919214777648449, -0.0038751689717173576, -0.004059332888573408, 0.0339323990046978, -0.016624143347144127, 0.06939277797937393, 0.0384554...
13
13
['Hao Wang', 'Dit-Yan Yeung']
1604.01662v2
While perception tasks such as visual object recognition and text understanding play an important role in human intelligence, the subsequent tasks that involve inference, reasoning and planning require an even higher level of intelligence. The past few years have seen major advances in many perception tasks using deep ...
Towards Bayesian Deep Learning: A Survey
2,016
http://arxiv.org/pdf/1604.01662v2
Title Towards Bayesian Deep Learning Survey Summary perception task visual object recognition text understanding play important role human intelligence subsequent task involve inference reasoning planning require even higher level intelligence past year seen major advance many perception task using deep learning model ...
[0.009172124788165092, 0.03827746957540512, 0.012193050235509872, 0.04512273892760277, -0.045876313000917435, 0.03563075140118599, 0.04301531985402107, 0.028921186923980713, -0.024594003334641457, -0.025386976078152657, -0.0037335555534809828, -0.013101317919790745, 0.035171762108802795, 0.07787491381168365, -0.0178403...
14
14
['Tejas D. Kulkarni', 'Karthik R. Narasimhan', 'Ardavan Saeedi', 'Joshua B. Tenenbaum']
1604.06057v2
Learning goal-directed behavior in environments with sparse feedback is a major challenge for reinforcement learning algorithms. The primary difficulty arises due to insufficient exploration, resulting in an agent being unable to learn robust value functions. Intrinsically motivated agents can explore new behavior for ...
Hierarchical Deep Reinforcement Learning: Integrating Temporal Abstraction and Intrinsic Motivation
2,016
http://arxiv.org/pdf/1604.06057v2
Title Hierarchical Deep Reinforcement Learning Integrating Temporal Abstraction Intrinsic Motivation Summary Learning goaldirected behavior environment sparse feedback major challenge reinforcement learning algorithm primary difficulty arises due insufficient exploration resulting agent unable learn robust value functi...
[0.007970927283167839, 0.09408263862133026, -0.010288913734257221, -0.03931441530585289, 0.028506653383374214, 0.0056037078611552715, -0.012039706110954285, -0.01628732867538929, -0.05267838016152382, 0.01272954884916544, -0.0309577826410532, 0.028655484318733215, -0.041666388511657715, 0.09871751815080643, 0.001510770...
15
15
['Deepak Pathak', 'Ross Girshick', 'Piotr Dollár', 'Trevor Darrell', 'Bharath Hariharan']
1612.06370v2
This paper presents a novel yet intuitive approach to unsupervised feature learning. Inspired by the human visual system, we explore whether low-level motion-based grouping cues can be used to learn an effective visual representation. Specifically, we use unsupervised motion-based segmentation on videos to obtain segme...
Learning Features by Watching Objects Move
2,016
http://arxiv.org/pdf/1612.06370v2
Title Learning Features Watching Objects Move Summary paper present novel yet intuitive approach unsupervised feature learning Inspired human visual system explore whether lowlevel motionbased grouping cue used learn effective visual representation Specifically use unsupervised motionbased segmentation video obtain seg...
[0.0038877902552485466, 0.009858185425400734, 0.002748560393229127, 0.047045670449733734, 0.016096679493784904, 0.03286033868789673, 0.030609730631113052, 0.0402018241584301, -0.04668239876627922, -0.03307424858212471, 0.007265650667250156, 0.01992909610271454, -0.019759509712457657, 0.03309622406959534, 0.018379300832...
16
16
['Muhammad Ghifary', 'W. Bastiaan Kleijn', 'Mengjie Zhang']
1409.6041v1
We propose a simple neural network model to deal with the domain adaptation problem in object recognition. Our model incorporates the Maximum Mean Discrepancy (MMD) measure as a regularization in the supervised learning to reduce the distribution mismatch between the source and target domains in the latent space. From ...
Domain Adaptive Neural Networks for Object Recognition
2,014
http://arxiv.org/pdf/1409.6041v1
Title Domain Adaptive Neural Networks Object Recognition Summary propose simple neural network model deal domain adaptation problem object recognition model incorporates Maximum Mean Discrepancy MMD measure regularization supervised learning reduce distribution mismatch source target domain latent space experiment demo...
[-0.014236598275601864, 0.008411363698542118, -0.018657177686691284, 0.0477081760764122, 0.019902747124433517, 0.014120462350547314, 0.05373150855302811, -0.015081859193742275, -0.037916991859674454, -0.01987382024526596, -0.015697956085205078, 0.010453196242451668, 0.007528562564402819, 0.047139428555965424, -0.000596...
17
17
['Lionel Pigou', 'Aäron van den Oord', 'Sander Dieleman', 'Mieke Van Herreweghe', 'Joni Dambre']
1506.01911v3
Recent studies have demonstrated the power of recurrent neural networks for machine translation, image captioning and speech recognition. For the task of capturing temporal structure in video, however, there still remain numerous open research questions. Current research suggests using a simple temporal feature pooling...
Beyond Temporal Pooling: Recurrence and Temporal Convolutions for Gesture Recognition in Video
2,015
http://arxiv.org/pdf/1506.01911v3
Title Beyond Temporal Pooling Recurrence Temporal Convolutions Gesture Recognition Video Summary Recent study demonstrated power recurrent neural network machine translation image captioning speech recognition task capturing temporal structure video however still remain numerous open research question Current research ...
[0.01242941152304411, 0.01989608444273472, 0.003044548910111189, 0.07898290455341339, 0.01224425807595253, 0.02652081660926342, 0.02579663135111332, 0.009492067620158195, -0.034116871654987335, -0.05787022411823273, -0.0010184143902733922, -0.06511043757200241, 0.05763678625226021, 0.03060113824903965, 0.01116096042096...
18
18
['Rakesh Achanta', 'Trevor Hastie']
1509.05962v2
In this paper, we address the task of Optical Character Recognition(OCR) for the Telugu script. We present an end-to-end framework that segments the text image, classifies the characters and extracts lines using a language model. The segmentation is based on mathematical morphology. The classification module, which is ...
Telugu OCR Framework using Deep Learning
2,015
http://arxiv.org/pdf/1509.05962v2
Title Telugu OCR Framework using Deep Learning Summary paper address task Optical Character RecognitionOCR Telugu script present endtoend framework segment text image classifies character extract line using language model segmentation based mathematical morphology classification module challenging task three deep convo...
[0.01340897660702467, 0.05425284430384636, 0.037259504199028015, 0.090110182762146, -0.0590578094124794, -0.0020389538258314133, 0.0018720458028838038, 0.04540619999170303, -0.017834406346082687, -0.003423208836466074, -0.000219842026126571, -0.03536640480160713, 0.048597030341625214, 0.06942123174667358, -0.0128559963...
19
19
['Jeff Donahue', 'Philipp Krähenbühl', 'Trevor Darrell']
1605.09782v7
The ability of the Generative Adversarial Networks (GANs) framework to learn generative models mapping from simple latent distributions to arbitrarily complex data distributions has been demonstrated empirically, with compelling results showing that the latent space of such generators captures semantic variation in the...
Adversarial Feature Learning
2,016
http://arxiv.org/pdf/1605.09782v7
Title Adversarial Feature Learning Summary ability Generative Adversarial Networks GANs framework learn generative model mapping simple latent distribution arbitrarily complex data distribution demonstrated empirically compelling result showing latent space generator capture semantic variation data distribution Intuiti...
[0.01773410104215145, 0.0888248085975647, -0.044434696435928345, 0.013442041352391243, -0.005503339692950249, 0.0025855363346636295, 0.004073440097272396, -0.015375635586678982, -0.03027379885315895, 0.00891395378857851, -0.0141568873077631, -0.0012658112682402134, -0.029098844155669212, 0.059320781379938126, 0.0601195...
20
20
['Zachary C. Lipton']
1606.03490v3
Supervised machine learning models boast remarkable predictive capabilities. But can you trust your model? Will it work in deployment? What else can it tell you about the world? We want models to be not only good, but interpretable. And yet the task of interpretation appears underspecified. Papers provide diverse and s...
The Mythos of Model Interpretability
2,016
http://arxiv.org/pdf/1606.03490v3
Title Mythos Model Interpretability Summary Supervised machine learning model boast remarkable predictive capability trust model work deployment else tell world want model good interpretable yet task interpretation appears underspecified Papers provide diverse sometimes nonoverlapping motivation interpretability offer ...
[0.01580430194735527, 0.03038216382265091, -0.02977801486849785, 0.018013877794146538, -0.001068954006768763, 0.02330859564244747, 0.024266917258501053, -0.000612442207057029, -0.034967511892318726, 0.012094943784177303, 0.011408815160393715, 0.03546974062919617, -0.006087920628488064, 0.10775260627269745, -0.003884392...
21
21
['Sahil Garg', 'Irina Rish', 'Guillermo Cecchi', 'Aurelie Lozano']
1701.06106v2
In this paper, we focus on online representation learning in non-stationary environments which may require continuous adaptation of model architecture. We propose a novel online dictionary-learning (sparse-coding) framework which incorporates the addition and deletion of hidden units (dictionary elements), and is inspi...
Neurogenesis-Inspired Dictionary Learning: Online Model Adaption in a Changing World
2,017
http://arxiv.org/pdf/1701.06106v2
Title NeurogenesisInspired Dictionary Learning Online Model Adaption Changing World Summary paper focus online representation learning nonstationary environment may require continuous adaptation model architecture propose novel online dictionarylearning sparsecoding framework incorporates addition deletion hidden unit ...
[-0.010662376880645752, 0.05564999580383301, -0.019077830016613007, -0.01993660070002079, 0.031982842832803726, 0.04325827211141586, 0.0516713447868824, 0.030131543055176735, -0.04375961422920227, 0.027563508599996567, 0.05267802253365517, -0.014716862700879574, -0.021795129403471947, 0.0833757221698761, 0.019378816708...
22
22
['Weifeng Ge', 'Yizhou Yu']
1702.08690v2
Deep neural networks require a large amount of labeled training data during supervised learning. However, collecting and labeling so much data might be infeasible in many cases. In this paper, we introduce a source-target selective joint fine-tuning scheme for improving the performance of deep learning tasks with insuf...
Borrowing Treasures from the Wealthy: Deep Transfer Learning through Selective Joint Fine-tuning
2,017
http://arxiv.org/pdf/1702.08690v2
Title Borrowing Treasures Wealthy Deep Transfer Learning Selective Joint Finetuning Summary Deep neural network require large amount labeled training data supervised learning However collecting labeling much data might infeasible many case paper introduce sourcetarget selective joint finetuning scheme improving perform...
[0.005758550483733416, 0.056478843092918396, -0.008075646124780178, 0.04061710089445114, 0.0007613872294314206, 0.003362262388691306, 0.0626605749130249, 0.01701948419213295, -0.04397859051823616, -0.00875970721244812, -0.04576868191361427, 0.04579819366335869, -0.011800899170339108, 0.04835931956768036, 0.013810884207...
23
23
['Tanmay Gupta', 'Kevin Shih', 'Saurabh Singh', 'Derek Hoiem']
1704.00260v2
An important goal of computer vision is to build systems that learn visual representations over time that can be applied to many tasks. In this paper, we investigate a vision-language embedding as a core representation and show that it leads to better cross-task transfer than standard multi-task learning. In particular...
Aligned Image-Word Representations Improve Inductive Transfer Across Vision-Language Tasks
2,017
http://arxiv.org/pdf/1704.00260v2
Title Aligned ImageWord Representations Improve Inductive Transfer Across VisionLanguage Tasks Summary important goal computer vision build system learn visual representation time applied many task paper investigate visionlanguage embedding core representation show lead better crosstask transfer standard multitask lear...
[0.014099230989813805, -0.02462138794362545, -0.009523318149149418, 0.08520937711000443, -0.01648499071598053, 0.02004041150212288, 0.00818576104938984, -0.007237681653350592, 0.020850639790296555, -0.055504005402326584, -0.05823567882180214, -0.010540205053985119, 0.011283772066235542, 0.026981983333826065, 0.04166373...
24
24
['Jan Hendrik Metzen', 'Mummadi Chaithanya Kumar', 'Thomas Brox', 'Volker Fischer']
1704.05712v3
While deep learning is remarkably successful on perceptual tasks, it was also shown to be vulnerable to adversarial perturbations of the input. These perturbations denote noise added to the input that was generated specifically to fool the system while being quasi-imperceptible for humans. More severely, there even exi...
Universal Adversarial Perturbations Against Semantic Image Segmentation
2,017
http://arxiv.org/pdf/1704.05712v3
Title Universal Adversarial Perturbations Semantic Image Segmentation Summary deep learning remarkably successful perceptual task also shown vulnerable adversarial perturbation input perturbation denote noise added input generated specifically fool system quasiimperceptible human severely even exist universal perturbat...
[0.0056486246176064014, 0.03561432287096977, -0.013600168749690056, 0.056583743542432785, 0.0028320043347775936, -0.005416540428996086, 0.0127531373873353, -0.02397274412214756, -0.04770808666944504, 0.009939886629581451, -0.01911712996661663, 0.06414027512073517, -0.027907874435186386, -0.019609153270721436, 0.0494938...
25
25
['Quynh Nguyen', 'Matthias Hein']
1704.08045v2
While the optimization problem behind deep neural networks is highly non-convex, it is frequently observed in practice that training deep networks seems possible without getting stuck in suboptimal points. It has been argued that this is the case as all local minima are close to being globally optimal. We show that thi...
The loss surface of deep and wide neural networks
2,017
http://arxiv.org/pdf/1704.08045v2
Title loss surface deep wide neural network Summary optimization problem behind deep neural network highly nonconvex frequently observed practice training deep network seems possible without getting stuck suboptimal point argued case local minimum close globally optimal show almost true fact almost local minimum global...
[-0.0181210245937109, 0.015704238787293434, -0.006439590826630592, 0.08110242336988449, -0.0018681740621104836, -0.01548056397587061, 0.04042204096913338, -0.020858725532889366, -0.04759621247649193, 0.03233534097671509, -0.010029901750385761, -0.0015024126041680574, -0.008791866712272167, 0.04432067275047302, 0.039892...
26
26
['Chris Donahue', 'Zachary C. Lipton', 'Akshay Balsubramani', 'Julian McAuley']
1705.07904v3
We propose a new algorithm for training generative adversarial networks that jointly learns latent codes for both identities (e.g. individual humans) and observations (e.g. specific photographs). By fixing the identity portion of the latent codes, we can generate diverse images of the same subject, and by fixing the ob...
Semantically Decomposing the Latent Spaces of Generative Adversarial Networks
2,017
http://arxiv.org/pdf/1705.07904v3
Title Semantically Decomposing Latent Spaces Generative Adversarial Networks Summary propose new algorithm training generative adversarial network jointly learns latent code identity eg individual human observation eg specific photograph fixing identity portion latent code generate diverse image subject fixing observat...
[0.02524767629802227, 0.1283176839351654, 0.0017059007659554482, 0.059948887676000595, -0.02689424343407154, 0.04963088035583496, 0.043878935277462006, -0.008901681751012802, -0.019719336181879044, 0.005113862454891205, -0.012964663095772266, -0.015912093222141266, 0.004586712457239628, 0.01671898551285267, 0.066325530...
27
27
['Mahesh Chandra Mukkamala', 'Matthias Hein']
1706.05507v2
Adaptive gradient methods have become recently very popular, in particular as they have been shown to be useful in the training of deep neural networks. In this paper we have analyzed RMSProp, originally proposed for the training of deep neural networks, in the context of online convex optimization and show $\sqrt{T}$-...
Variants of RMSProp and Adagrad with Logarithmic Regret Bounds
2,017
http://arxiv.org/pdf/1706.05507v2
Title Variants RMSProp Adagrad Logarithmic Regret Bounds Summary Adaptive gradient method become recently popular particular shown useful training deep neural network paper analyzed RMSProp originally proposed training deep neural network context online convex optimization show sqrtTtype regret bound Moreover propose t...
[-0.0235554501414299, 0.01657301001250744, 0.022070230916142464, -0.00982610508799553, 0.021564722061157227, -0.01518192794173956, 0.008455047383904457, -0.03350786492228508, -0.05014415457844734, 0.018914753571152687, -0.045846689492464066, -0.016145600005984306, 0.004574382212013006, 0.003473706543445587, 0.011971698...
28
28
['Chunyuan Li', 'Hao Liu', 'Changyou Chen', 'Yunchen Pu', 'Liqun Chen', 'Ricardo Henao', 'Lawrence Carin']
1709.01215v2
We investigate the non-identifiability issues associated with bidirectional adversarial training for joint distribution matching. Within a framework of conditional entropy, we propose both adversarial and non-adversarial approaches to learn desirable matched joint distributions for unsupervised and supervised tasks. We...
ALICE: Towards Understanding Adversarial Learning for Joint Distribution Matching
2,017
http://arxiv.org/pdf/1709.01215v2
Title ALICE Towards Understanding Adversarial Learning Joint Distribution Matching Summary investigate nonidentifiability issue associated bidirectional adversarial training joint distribution matching Within framework conditional entropy propose adversarial nonadversarial approach learn desirable matched joint distrib...
[0.006879579741507769, 0.07195214182138443, -0.011232499964535236, 0.04159289970993996, -0.024009624496102333, -0.016379093751311302, 0.016849802806973457, 0.004335680510848761, 0.012014094740152359, -0.032395802438259125, -0.05185306817293167, -0.0074209896847605705, -0.010995801538228989, 0.018014095723628998, 0.0075...
29
29
['Mateusz Buda', 'Atsuto Maki', 'Maciej A. Mazurowski']
1710.05381v1
In this study, we systematically investigate the impact of class imbalance on classification performance of convolutional neural networks (CNNs) and compare frequently used methods to address the issue. Class imbalance is a common problem that has been comprehensively studied in classical machine learning, yet very lim...
A systematic study of the class imbalance problem in convolutional neural networks
2,017
http://arxiv.org/pdf/1710.05381v1
Title systematic study class imbalance problem convolutional neural network Summary study systematically investigate impact class imbalance classification performance convolutional neural network CNNs compare frequently used method address issue Class imbalance common problem comprehensively studied classical machine l...
[0.002548755845054984, 0.014575646258890629, -0.052617598325014114, 0.008142157457768917, 0.019604474306106567, -0.008515378460288048, 0.06046532839536667, -0.010335146449506283, -0.04602038115262985, -0.041656531393527985, 0.00910205114632845, 0.044474340975284576, 0.00974102783948183, 0.022866446524858475, -0.0017808...
30
30
['Jan Kukačka', 'Vladimir Golkov', 'Daniel Cremers']
1710.10686v1
Regularization is one of the crucial ingredients of deep learning, yet the term regularization has various definitions, and regularization methods are often studied separately from each other. In our work we present a systematic, unifying taxonomy to categorize existing methods. We distinguish methods that affect data,...
Regularization for Deep Learning: A Taxonomy
2,017
http://arxiv.org/pdf/1710.10686v1
Title Regularization Deep Learning Taxonomy Summary Regularization one crucial ingredient deep learning yet term regularization various definition regularization method often studied separately work present systematic unifying taxonomy categorize existing method distinguish method affect data network architecture error...
[0.0039631095714867115, 0.04202922433614731, -0.029790757223963737, 0.013474303297698498, 0.0033911168575286865, -0.035869937390089035, 0.06577468663454056, 0.028167083859443665, -0.05148724839091301, 0.006152178626507521, -0.003993260208517313, -0.030590925365686417, 0.019640978425741196, 0.05663751810789108, 0.006232...
31
31
['Elie Aljalbout', 'Vladimir Golkov', 'Yawar Siddiqui', 'Daniel Cremers']
1801.07648v1
Clustering is a fundamental machine learning method. The quality of its results is dependent on the data distribution. For this reason, deep neural networks can be used for learning better representations of the data. In this paper, we propose a systematic taxonomy for clustering with deep learning, in addition to a re...
Clustering with Deep Learning: Taxonomy and New Methods
2,018
http://arxiv.org/pdf/1801.07648v1
Title Clustering Deep Learning Taxonomy New Methods Summary Clustering fundamental machine learning method quality result dependent data distribution reason deep neural network used learning better representation data paper propose systematic taxonomy clustering deep learning addition review method field Based taxonomy...
[-0.016683556139469147, 0.010083289816975594, -0.03315931186079979, 0.029727904126048088, 0.009991924278438091, -0.005651251878589392, 0.05703303590416908, -0.00812006276100874, -0.0044061243534088135, -0.00665407907217741, -0.03286062553524971, -0.004478133749216795, 0.025629807263612747, 0.059234000742435455, 0.01990...
32
32
['Armand Zampieri', 'Guillaume Charpiat', 'Yuliya Tarabalka']
1802.09816v1
We tackle here the problem of multimodal image non-rigid registration, which is of prime importance in remote sensing and medical imaging. The difficulties encountered by classical registration approaches include feature design and slow optimization by gradient descent. By analyzing these methods, we note the significa...
Coarse to fine non-rigid registration: a chain of scale-specific neural networks for multimodal image alignment with application to remote sensing
2,018
http://arxiv.org/pdf/1802.09816v1
Title Coarse fine nonrigid registration chain scalespecific neural network multimodal image alignment application remote sensing Summary tackle problem multimodal image nonrigid registration prime importance remote sensing medical imaging difficulty encountered classical registration approach include feature design slo...
[-0.0002296001766808331, 0.027488330379128456, 0.0224667489528656, 0.02791258506476879, -0.05824168026447296, -0.00564225297421217, 0.019202496856451035, -0.06031869351863861, -0.022075379267334938, 0.027093613520264626, 0.028951166197657585, -0.004241004586219788, 0.04826965183019638, 0.03934275731444359, 0.0270466450...
33
33
['Li Yao', 'Atousa Torabi', 'Kyunghyun Cho', 'Nicolas Ballas', 'Christopher Pal', 'Hugo Larochelle', 'Aaron Courville']
1502.08029v5
Recent progress in using recurrent neural networks (RNNs) for image description has motivated the exploration of their application for video description. However, while images are static, working with videos requires modeling their dynamic temporal structure and then properly integrating that information into a natural...
Describing Videos by Exploiting Temporal Structure
2,015
http://arxiv.org/pdf/1502.08029v5
Title Describing Videos Exploiting Temporal Structure Summary Recent progress using recurrent neural network RNNs image description motivated exploration application video description However image static working video requires modeling dynamic temporal structure properly integrating information natural language descri...
[0.04384573921561241, 0.040353644639253616, 0.022152865305542946, 0.06905457377433777, -0.0016936935717239976, 0.004977921023964882, -0.015876099467277527, -0.016332415863871574, -0.07429931312799454, -0.07366855442523956, 0.010147901251912117, -0.06954325735569, 0.04226887226104736, 0.06193795055150986, 0.006885376758...
34
34
['Hao Wang', 'Xingjian Shi', 'Dit-Yan Yeung']
1611.00454v1
Hybrid methods that utilize both content and rating information are commonly used in many recommender systems. However, most of them use either handcrafted features or the bag-of-words representation as a surrogate for the content information but they are neither effective nor natural enough. To address this problem, w...
Collaborative Recurrent Autoencoder: Recommend while Learning to Fill in the Blanks
2,016
http://arxiv.org/pdf/1611.00454v1
Title Collaborative Recurrent Autoencoder Recommend Learning Fill Blanks Summary Hybrid method utilize content rating information commonly used many recommender system However use either handcrafted feature bagofwords representation surrogate content information neither effective natural enough address problem develop ...
[0.026162119582295418, 0.03550722077488899, 0.0006124278297647834, 0.02112296223640442, -0.0031149424612522125, 0.0052260784432291985, 0.03841988742351532, 0.014339396730065346, -0.011528442613780499, -0.0230964794754982, -0.04469263553619385, -0.018563508987426758, 0.006120844278484583, 0.09863444417715073, -0.0506608...
35
35
['Laura Graesser', 'Abhinav Gupta', 'Lakshay Sharma', 'Evelina Bakhturina']
1712.00725v1
In this project we analysed how much semantic information images carry, and how much value image data can add to sentiment analysis of the text associated with the images. To better understand the contribution from images, we compared models which only made use of image data, models which only made use of text data, an...
Sentiment Classification using Images and Label Embeddings
2,017
http://arxiv.org/pdf/1712.00725v1
Title Sentiment Classification using Images Label Embeddings Summary project analysed much semantic information image carry much value image data add sentiment analysis text associated image better understand contribution image compared model made use image data model made use text data model combined data type also an...
[0.0373409278690815, 0.08658852428197861, 0.0031289902981370687, 0.08435819298028946, -0.022962214425206184, 0.04876367002725601, -0.011424457654356956, 0.010839671827852726, 0.024760400876402855, -0.06766130030155182, -0.022244643419981003, 0.03267093002796173, -0.010490966029465199, 0.03902794048190117, 0.00771614862...
36
36
['Hao Wang', 'Xingjian Shi', 'Dit-Yan Yeung']
1611.00448v1
Neural networks (NN) have achieved state-of-the-art performance in various applications. Unfortunately in applications where training data is insufficient, they are often prone to overfitting. One effective way to alleviate this problem is to exploit the Bayesian approach by using Bayesian neural networks (BNN). Anothe...
Natural-Parameter Networks: A Class of Probabilistic Neural Networks
2,016
http://arxiv.org/pdf/1611.00448v1
Title NaturalParameter Networks Class Probabilistic Neural Networks Summary Neural network NN achieved stateoftheart performance various application Unfortunately application training data insufficient often prone overfitting One effective way alleviate problem exploit Bayesian approach using Bayesian neural network BN...
[-0.019308457151055336, 0.04892473667860031, -0.018579134717583656, -0.011225856840610504, -0.007168007083237171, -0.058175358921289444, 0.012418322265148163, -0.012458806857466698, -0.029085859656333923, 0.007147525902837515, -0.000888597802259028, 0.04558419808745384, 0.017681701108813286, 0.04631771147251129, 0.0518...
37
37
['Misha Denil', 'Pulkit Agrawal', 'Tejas D Kulkarni', 'Tom Erez', 'Peter Battaglia', 'Nando de Freitas']
1611.01843v3
When encountering novel objects, humans are able to infer a wide range of physical properties such as mass, friction and deformability by interacting with them in a goal driven way. This process of active interaction is in the same spirit as a scientist performing experiments to discover hidden facts. Recent advances i...
Learning to Perform Physics Experiments via Deep Reinforcement Learning
2,016
http://arxiv.org/pdf/1611.01843v3
Title Learning Perform Physics Experiments via Deep Reinforcement Learning Summary encountering novel object human able infer wide range physical property mass friction deformability interacting goal driven way process active interaction spirit scientist performing experiment discover hidden fact Recent advance artific...
[0.004888972733169794, 0.022142047062516212, -0.025890188291668892, 0.007126957178115845, -0.003003097604960203, -0.02418522723019123, 0.06944964826107025, 0.01833062805235386, -0.04840927571058273, 0.04805910214781761, 0.003001939970999956, 0.03375035524368286, -0.037421874701976776, 0.11026794463396072, 0.01542024966...
38
38
['Tsung-Hsien Wen', 'David Vandyke', 'Nikola Mrksic', 'Milica Gasic', 'Lina M. Rojas-Barahona', 'Pei-Hao Su', 'Stefan Ultes', 'Steve Young']
1604.04562v3
Teaching machines to accomplish tasks by conversing naturally with humans is challenging. Currently, developing task-oriented dialogue systems requires creating multiple components and typically this involves either a large amount of handcrafting, or acquiring costly labelled datasets to solve a statistical learning pr...
A Network-based End-to-End Trainable Task-oriented Dialogue System
2,016
http://arxiv.org/pdf/1604.04562v3
Title Networkbased EndtoEnd Trainable Taskoriented Dialogue System Summary Teaching machine accomplish task conversing naturally human challenging Currently developing taskoriented dialogue system requires creating multiple component typically involves either large amount handcrafting acquiring costly labelled datasets...
[0.05435043200850487, 0.017975132912397385, -0.004753083921968937, 0.06381011754274368, -0.01053185947239399, 0.012891801074147224, 0.013492371886968613, -0.020046189427375793, 0.006221742369234562, -0.04255888611078262, -0.04956076666712761, -0.00783568900078535, 0.007973398081958294, 0.07707810401916504, 0.0171229392...
39
39
['Johannes Welbl', 'Guillaume Bouchard', 'Sebastian Riedel']
1604.05878v1
Embedding-based Knowledge Base Completion models have so far mostly combined distributed representations of individual entities or relations to compute truth scores of missing links. Facts can however also be represented using pairwise embeddings, i.e. embeddings for pairs of entities and relations. In this paper we ex...
A Factorization Machine Framework for Testing Bigram Embeddings in Knowledgebase Completion
2,016
http://arxiv.org/pdf/1604.05878v1
Title Factorization Machine Framework Testing Bigram Embeddings Knowledgebase Completion Summary Embeddingbased Knowledge Base Completion model far mostly combined distributed representation individual entity relation compute truth score missing link Facts however also represented using pairwise embeddings ie embedding...
[0.000826517993118614, 0.009025746956467628, -0.005209268070757389, 0.035159654915332794, 0.0032346744555979967, 0.027672480791807175, -0.006383709143847227, -0.011465386487543583, 0.03268832713365555, -0.02079460769891739, 0.02257765643298626, 0.025877708569169044, -0.0053927102126181126, 0.02905987948179245, -0.00385...
40
40
['Franck Dernoncourt', 'Ji Young Lee', 'Peter Szolovits']
1612.05251v1
Existing models based on artificial neural networks (ANNs) for sentence classification often do not incorporate the context in which sentences appear, and classify sentences individually. However, traditional sentence classification approaches have been shown to greatly benefit from jointly classifying subsequent sente...
Neural Networks for Joint Sentence Classification in Medical Paper Abstracts
2,016
http://arxiv.org/pdf/1612.05251v1
Title Neural Networks Joint Sentence Classification Medical Paper Abstracts Summary Existing model based artificial neural network ANNs sentence classification often incorporate context sentence appear classify sentence individually However traditional sentence classification approach shown greatly benefit jointly clas...
[0.04733778536319733, 0.0364992655813694, -0.0035521562676876783, 0.002981181489303708, -0.042058493942022324, 0.03678615391254425, 0.025012049823999405, 0.014855986461043358, 0.01904723420739174, -0.04209323227405548, -0.02876879833638668, -0.05620887130498886, 0.036804135888814926, -0.0016750863287597895, -0.00407067...
41
41
['Franck Dernoncourt', 'Ji Young Lee', 'Ozlem Uzuner', 'Peter Szolovits']
1606.03475v1
Objective: Patient notes in electronic health records (EHRs) may contain critical information for medical investigations. However, the vast majority of medical investigators can only access de-identified notes, in order to protect the confidentiality of patients. In the United States, the Health Insurance Portability a...
De-identification of Patient Notes with Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1606.03475v1
Title Deidentification Patient Notes Recurrent Neural Networks Summary Objective Patient note electronic health record EHRs may contain critical information medical investigation However vast majority medical investigator access deidentified note order protect confidentiality patient United States Health Insurance Port...
[0.021877270191907883, 0.06426245719194412, -0.01817529834806919, -0.019433828070759773, -0.002087124390527606, 0.019773103296756744, 0.027331892400979996, 0.010316393338143826, 0.004598485771566629, -0.004994913004338741, 0.04128994792699814, -0.01306468341499567, 0.03758978098630905, 0.048275794833898544, -0.01015557...
42
42
['Tsendsuren Munkhdalai', 'Hong Yu']
1610.06454v2
Hypothesis testing is an important cognitive process that supports human reasoning. In this paper, we introduce a computational hypothesis testing approach based on memory augmented neural networks. Our approach involves a hypothesis testing loop that reconsiders and progressively refines a previously formed hypothesis...
Reasoning with Memory Augmented Neural Networks for Language Comprehension
2,016
http://arxiv.org/pdf/1610.06454v2
Title Reasoning Memory Augmented Neural Networks Language Comprehension Summary Hypothesis testing important cognitive process support human reasoning paper introduce computational hypothesis testing approach based memory augmented neural network approach involves hypothesis testing loop reconsiders progressively refin...
[0.03371580317616463, -0.004798790905624628, -0.01564905047416687, 0.056556738913059235, -0.03314266726374626, 0.025938093662261963, 0.02255520410835743, -0.03362160921096802, -0.0031821217853575945, -0.017369162291288376, 0.031453315168619156, -0.028457975015044212, 0.03600015118718147, 0.0016937246546149254, 0.024398...
43
43
['W. James Murdoch', 'Arthur Szlam']
1702.02540v2
Although deep learning models have proven effective at solving problems in natural language processing, the mechanism by which they come to their conclusions is often unclear. As a result, these models are generally treated as black boxes, yielding no insight of the underlying learned patterns. In this paper we conside...
Automatic Rule Extraction from Long Short Term Memory Networks
2,017
http://arxiv.org/pdf/1702.02540v2
Title Automatic Rule Extraction Long Short Term Memory Networks Summary Although deep learning model proven effective solving problem natural language processing mechanism come conclusion often unclear result model generally treated black box yielding insight underlying learned pattern paper consider Long Short Term Me...
[0.05388208106160164, -0.0057205636985599995, -0.008655648678541183, 0.07049400359392166, -0.06092309206724167, 0.0017530766781419516, -0.010157352313399315, 0.040230054408311844, 0.0032108710147440434, -0.08655161410570145, 0.028754491358995438, -0.014085361734032631, -0.01891648955643177, 0.06222238019108772, -0.0232...
44
44
['Sebastian Gehrmann', 'Franck Dernoncourt', 'Yeran Li', 'Eric T. Carlson', 'Joy T. Wu', 'Jonathan Welt', 'John Foote Jr.', 'Edward T. Moseley', 'David W. Grant', 'Patrick D. Tyler', 'Leo Anthony Celi']
1703.08705v1
Objective: We investigate whether deep learning techniques for natural language processing (NLP) can be used efficiently for patient phenotyping. Patient phenotyping is a classification task for determining whether a patient has a medical condition, and is a crucial part of secondary analysis of healthcare data. We ass...
Comparing Rule-Based and Deep Learning Models for Patient Phenotyping
2,017
http://arxiv.org/pdf/1703.08705v1
Title Comparing RuleBased Deep Learning Models Patient Phenotyping Summary Objective investigate whether deep learning technique natural language processing NLP used efficiently patient phenotyping Patient phenotyping classification task determining whether patient medical condition crucial part secondary analysis heal...
[0.05393210053443909, 0.017177172005176544, 0.003052765503525734, -0.008947459980845451, -0.010735324583947659, 0.01852540113031864, 0.01480336394160986, 0.013665424659848213, -0.02276225946843624, -0.022496294230222702, 0.010864540003240108, -0.05141543224453926, 0.015303588472306728, 0.06716867536306381, 0.0016145890...
45
45
['Ji Young Lee', 'Franck Dernoncourt', 'Peter Szolovits']
1704.01523v1
Over 50 million scholarly articles have been published: they constitute a unique repository of knowledge. In particular, one may infer from them relations between scientific concepts, such as synonyms and hyponyms. Artificial neural networks have been recently explored for relation extraction. In this work, we continue...
MIT at SemEval-2017 Task 10: Relation Extraction with Convolutional Neural Networks
2,017
http://arxiv.org/pdf/1704.01523v1
Title MIT SemEval2017 Task 10 Relation Extraction Convolutional Neural Networks Summary 50 million scholarly article published constitute unique repository knowledge particular one may infer relation scientific concept synonym hyponym Artificial neural network recently explored relation extraction work continue line wo...
[0.058312851935625076, 0.06265062093734741, 0.005281275138258934, 0.055801503360271454, -0.031507428735494614, -0.00939155276864767, 0.026195526123046875, 0.029269294813275337, -0.028874466195702553, -0.037247151136398315, -0.02918235957622528, 0.05256698280572891, 0.0099489139392972, 0.0036105012986809015, -0.01692311...
46
46
['Ji Young Lee', 'Franck Dernoncourt', 'Peter Szolovits']
1705.06273v1
Recent approaches based on artificial neural networks (ANNs) have shown promising results for named-entity recognition (NER). In order to achieve high performances, ANNs need to be trained on a large labeled dataset. However, labels might be difficult to obtain for the dataset on which the user wants to perform NER: la...
Transfer Learning for Named-Entity Recognition with Neural Networks
2,017
http://arxiv.org/pdf/1705.06273v1
Title Transfer Learning NamedEntity Recognition Neural Networks Summary Recent approach based artificial neural network ANNs shown promising result namedentity recognition NER order achieve high performance ANNs need trained large labeled dataset However label might difficult obtain dataset user want perform NER label ...
[0.03788178414106369, 0.0226596649736166, 0.0063038170337677, 0.0070561072789132595, 0.0062925005331635475, 0.019892117008566856, 0.005912716966122389, 0.012715619057416916, -0.0196891687810421, -0.0015258564380928874, -0.04504218325018883, -0.004810972139239311, 0.02148493379354477, 0.0035197273828089237, -0.013179363...
47
47
['Sai Rajeswar', 'Sandeep Subramanian', 'Francis Dutil', 'Christopher Pal', 'Aaron Courville']
1705.10929v1
Generative Adversarial Networks (GANs) have gathered a lot of attention from the computer vision community, yielding impressive results for image generation. Advances in the adversarial generation of natural language from noise however are not commensurate with the progress made in generating images, and still lag far ...
Adversarial Generation of Natural Language
2,017
http://arxiv.org/pdf/1705.10929v1
Title Adversarial Generation Natural Language Summary Generative Adversarial Networks GANs gathered lot attention computer vision community yielding impressive result image generation Advances adversarial generation natural language noise however commensurate progress made generating image still lag far behind likeliho...
[0.05477595701813698, 0.08053953945636749, -0.013415508903563023, 0.04601915925741196, -0.016717227175831795, -0.011133761145174503, 0.01022478099912405, -0.007239060942083597, 0.0230791587382555, -0.03693745657801628, 0.005420372821390629, -0.01803283952176571, 0.0035774034913629293, 0.03436025604605675, 0.07447671145...
48
48
['Leila Arras', 'Grégoire Montavon', 'Klaus-Robert Müller', 'Wojciech Samek']
1706.07206v2
Recently, a technique called Layer-wise Relevance Propagation (LRP) was shown to deliver insightful explanations in the form of input space relevances for understanding feed-forward neural network classification decisions. In the present work, we extend the usage of LRP to recurrent neural networks. We propose a specif...
Explaining Recurrent Neural Network Predictions in Sentiment Analysis
2,017
http://arxiv.org/pdf/1706.07206v2
Title Explaining Recurrent Neural Network Predictions Sentiment Analysis Summary Recently technique called Layerwise Relevance Propagation LRP shown deliver insightful explanation form input space relevance understanding feedforward neural network classification decision present work extend usage LRP recurrent neural n...
[0.04484899342060089, 0.02378677763044834, -0.00305969943292439, 0.051354214549064636, -0.011041464284062386, -0.0319642499089241, -0.016614774242043495, -0.006575544364750385, -0.019484154880046844, -0.045627813786268234, -0.01730375364422798, -0.010188334621489048, 0.0017385354731231928, 0.025300318375229836, -0.0215...
49
49
['Emmanuel Dufourq', 'Bruce A. Bassett']
1709.06990v1
Can textual data be compressed intelligently without losing accuracy in evaluating sentiment? In this study, we propose a novel evolutionary compression algorithm, PARSEC (PARts-of-Speech for sEntiment Compression), which makes use of Parts-of-Speech tags to compress text in a way that sacrifices minimal classification...
Text Compression for Sentiment Analysis via Evolutionary Algorithms
2,017
http://arxiv.org/pdf/1709.06990v1
Title Text Compression Sentiment Analysis via Evolutionary Algorithms Summary textual data compressed intelligently without losing accuracy evaluating sentiment study propose novel evolutionary compression algorithm PARSEC PARtsofSpeech sEntiment Compression make use PartsofSpeech tag compress text way sacrifice minima...
[0.029672643169760704, 0.0624702051281929, -0.022342588752508163, 0.011170751415193081, -0.05197097361087799, -0.0025711855851113796, -0.06413792073726654, 0.06467556208372116, -0.019226262345910072, -0.0199278611689806, 0.03828226029872894, 0.02604842372238636, 0.011373891495168209, 0.04111306369304657, -0.03516996279...
50
50
['Kartik Audhkhasi', 'Brian Kingsbury', 'Bhuvana Ramabhadran', 'George Saon', 'Michael Picheny']
1712.03133v1
Direct acoustics-to-word (A2W) models in the end-to-end paradigm have received increasing attention compared to conventional sub-word based automatic speech recognition models using phones, characters, or context-dependent hidden Markov model states. This is because A2W models recognize words from speech without any de...
Building competitive direct acoustics-to-word models for English conversational speech recognition
2,017
http://arxiv.org/pdf/1712.03133v1
Title Building competitive direct acousticstoword model English conversational speech recognition Summary Direct acousticstoword A2W model endtoend paradigm received increasing attention compared conventional subword based automatic speech recognition model using phone character contextdependent hidden Markov model sta...
[0.04574203118681908, 0.025533463805913925, 0.042400166392326355, 0.04855414852499962, -0.05226854234933853, -0.0042722481302917, 0.020178286358714104, 0.024636832997202873, -0.015204379335045815, -0.06854008883237839, -0.0555407777428627, -0.024402471259236336, 0.05278247594833374, 0.013023776933550835, 0.005281993187...
51
51
['Huijuan Xu', 'Kate Saenko']
1511.05234v2
We address the problem of Visual Question Answering (VQA), which requires joint image and language understanding to answer a question about a given photograph. Recent approaches have applied deep image captioning methods based on convolutional-recurrent networks to this problem, but have failed to model spatial inferen...
Ask, Attend and Answer: Exploring Question-Guided Spatial Attention for Visual Question Answering
2,015
http://arxiv.org/pdf/1511.05234v2
Title Ask Attend Answer Exploring QuestionGuided Spatial Attention Visual Question Answering Summary address problem Visual Question Answering VQA requires joint image language understanding answer question given photograph Recent approach applied deep image captioning method based convolutionalrecurrent network proble...
[0.0524277538061142, 0.029206106439232826, -0.024834005162119865, 0.04498821869492531, -0.006906989496201277, 0.009571080096065998, 0.025222986936569214, -0.0028687838930636644, 0.02206416241824627, -0.03178846836090088, 0.0006940392195247114, 0.005100044887512922, -0.0010749456705525517, 0.04998240992426872, 0.0496853...
52
52
['Yuetan Lin', 'Zhangyang Pang', 'Donghui Wang', 'Yueting Zhuang']
1702.06700v1
Visual question answering (VQA) has witnessed great progress since May, 2015 as a classic problem unifying visual and textual data into a system. Many enlightening VQA works explore deep into the image and question encodings and fusing methods, of which attention is the most effective and infusive mechanism. Current at...
Task-driven Visual Saliency and Attention-based Visual Question Answering
2,017
http://arxiv.org/pdf/1702.06700v1
Title Taskdriven Visual Saliency Attentionbased Visual Question Answering Summary Visual question answering VQA witnessed great progress since May 2015 classic problem unifying visual textual data system Many enlightening VQA work explore deep image question encoding fusing method attention effective infusive mechanism...
[0.03881014138460159, -0.005521686282008886, -0.0014341219794005156, 0.07670561969280243, 0.02064802125096321, 0.014323089271783829, -0.014976751990616322, 0.008460894227027893, -0.003425849135965109, -0.02413828670978546, -0.009696214459836483, -0.00717691658064723, -0.0037670289166271687, 0.053044453263282776, 0.0465...
53
53
['Akash Kumar Dhaka', 'Giampiero Salvi']
1606.09163v1
We present a systematic analysis on the performance of a phonetic recogniser when the window of input features is not symmetric with respect to the current frame. The recogniser is based on Context Dependent Deep Neural Networks (CD-DNNs) and Hidden Markov Models (HMMs). The objective is to reduce the latency of the sy...
Optimising The Input Window Alignment in CD-DNN Based Phoneme Recognition for Low Latency Processing
2,016
http://arxiv.org/pdf/1606.09163v1
Title Optimising Input Window Alignment CDDNN Based Phoneme Recognition Low Latency Processing Summary present systematic analysis performance phonetic recogniser window input feature symmetric respect current frame recogniser based Context Dependent Deep Neural Networks CDDNNs Hidden Markov Models HMMs objective reduc...
[-0.020944718271493912, 0.029016690328717232, 0.008884217590093613, 0.02872476726770401, 0.004271514713764191, -0.018700644373893738, 0.07468104362487793, 0.004588223993778229, -0.027082746848464012, 0.040426988154649734, -0.06507018953561783, -0.042696014046669006, 0.08818154036998749, 0.04114234820008278, 0.022322118...
54
54
['Peng Qian', 'Xipeng Qiu', 'Xuanjing Huang']
1604.06635v1
Recently, the long short-term memory neural network (LSTM) has attracted wide interest due to its success in many tasks. LSTM architecture consists of a memory cell and three gates, which looks similar to the neuronal networks in the brain. However, there still lacks the evidence of the cognitive plausibility of LSTM a...
Bridging LSTM Architecture and the Neural Dynamics during Reading
2,016
http://arxiv.org/pdf/1604.06635v1
Title Bridging LSTM Architecture Neural Dynamics Reading Summary Recently long shortterm memory neural network LSTM attracted wide interest due success many task LSTM architecture consists memory cell three gate look similar neuronal network brain However still lack evidence cognitive plausibility LSTM architecture wel...
[-0.012993956916034222, -0.006285125855356455, -0.015956982970237732, 0.033215660601854324, 0.0012327719014137983, 0.015472170896828175, 0.039305780082941055, 0.008511208929121494, 0.027965713292360306, 0.04192258045077324, -0.06797993928194046, -0.034531183540821075, 0.0488734170794487, 0.05301915854215622, 0.05934487...
55
55
['Jiwei Li']
1412.3714v2
This paper addresses how a recursive neural network model can automatically leave out useless information and emphasize important evidence, in other words, to perform "weight tuning" for higher-level representation acquisition. We propose two models, Weighted Neural Network (WNN) and Binary-Expectation Neural Network (...
Feature Weight Tuning for Recursive Neural Networks
2,014
http://arxiv.org/pdf/1412.3714v2
Title Feature Weight Tuning Recursive Neural Networks Summary paper address recursive neural network model automatically leave useless information emphasize important evidence word perform weight tuning higherlevel representation acquisition propose two model Weighted Neural Network WNN BinaryExpectation Neural Network...
[0.012952763587236404, 0.029821977019309998, -0.01966494508087635, 0.03055460937321186, 0.0001618549576960504, -0.003271478693932295, 0.0013514960883185267, -0.005855271592736244, -0.020848331972956657, 0.002556001301854849, 0.03033561259508133, -0.017845505848526955, 0.03005438670516014, 0.051921360194683075, 0.006054...
56
56
['Sadikin Mujiono', 'Mohamad Ivan Fanany', 'Chan Basaruddin']
1610.01891v1
One essential task in information extraction from the medical corpus is drug name recognition. Compared with text sources come from other domains, the medical text is special and has unique characteristics. In addition, the medical text mining poses more challenges, e.g., more unstructured text, the fast growing of new...
A New Data Representation Based on Training Data Characteristics to Extract Drug Named-Entity in Medical Text
2,016
http://arxiv.org/pdf/1610.01891v1
Title New Data Representation Based Training Data Characteristics Extract Drug NamedEntity Medical Text Summary One essential task information extraction medical corpus drug name recognition Compared text source come domain medical text special unique characteristic addition medical text mining pose challenge eg unstru...
[0.033277690410614014, -0.004981243517249823, 0.00892974715679884, 0.012650229968130589, -0.012423137202858925, 0.005561047233641148, 0.0060630664229393005, 0.017601408064365387, 0.005051193293184042, -0.04098929092288017, -0.026254603639245033, -0.01484074629843235, -0.007727933581918478, 0.06162698566913605, 0.017386...
57
57
['Eric Malmi', 'Pyry Takala', 'Hannu Toivonen', 'Tapani Raiko', 'Aristides Gionis']
1505.04771v2
Writing rap lyrics requires both creativity to construct a meaningful, interesting story and lyrical skills to produce complex rhyme patterns, which form the cornerstone of good flow. We present a rap lyrics generation method that captures both of these aspects. First, we develop a prediction model to identify the next...
DopeLearning: A Computational Approach to Rap Lyrics Generation
2,015
http://arxiv.org/pdf/1505.04771v2
Title DopeLearning Computational Approach Rap Lyrics Generation Summary Writing rap lyric requires creativity construct meaningful interesting story lyrical skill produce complex rhyme pattern form cornerstone good flow present rap lyric generation method capture aspect First develop prediction model identify next line...
[0.06822919845581055, 0.032081488519907, -0.01135606225579977, -0.01575624570250511, -0.04941854253411293, -0.0047858408652246, 0.006555943284183741, -0.016065383329987526, -0.03747532144188881, -0.009168005548417568, -0.0009104200289584696, 0.01905684545636177, 0.0540979728102684, 0.021996311843395233, 0.0182785503566...
58
58
['Shengxian Wan', 'Yanyan Lan', 'Jun Xu', 'Jiafeng Guo', 'Liang Pang', 'Xueqi Cheng']
1604.04378v1
Semantic matching, which aims to determine the matching degree between two texts, is a fundamental problem for many NLP applications. Recently, deep learning approach has been applied to this problem and significant improvements have been achieved. In this paper, we propose to view the generation of the global interact...
Match-SRNN: Modeling the Recursive Matching Structure with Spatial RNN
2,016
http://arxiv.org/pdf/1604.04378v1
Title MatchSRNN Modeling Recursive Matching Structure Spatial RNN Summary Semantic matching aim determine matching degree two text fundamental problem many NLP application Recently deep learning approach applied problem significant improvement achieved paper propose view generation global interaction two text recursive...
[0.04786315932869911, 0.031401198357343674, -0.015540102496743202, 0.07065323740243912, -0.04607900232076645, -0.009140062145888805, -0.019094157963991165, 0.0045071071945130825, -0.009520016610622406, -0.014485950581729412, 0.0053553651086986065, -0.022471679374575615, 0.04384959116578102, 0.06489219516515732, -0.0041...
59
59
['Iulian V. Serban', 'Alexander G. Ororbia II', 'Joelle Pineau', 'Aaron Courville']
1612.00377v4
Advances in neural variational inference have facilitated the learning of powerful directed graphical models with continuous latent variables, such as variational autoencoders. The hope is that such models will learn to represent rich, multi-modal latent factors in real-world data, such as natural language text. Howeve...
Piecewise Latent Variables for Neural Variational Text Processing
2,016
http://arxiv.org/pdf/1612.00377v4
Title Piecewise Latent Variables Neural Variational Text Processing Summary Advances neural variational inference facilitated learning powerful directed graphical model continuous latent variable variational autoencoders hope model learn represent rich multimodal latent factor realworld data natural language text Howev...
[0.05198274552822113, 0.09823858737945557, -0.033034391701221466, 0.02786707691848278, -0.02450750023126602, 0.009729685261845589, 0.03232638165354729, -0.0357762835919857, -0.05808180943131447, -0.04140713810920715, 0.031163692474365234, -0.05717034265398979, -0.014125913381576538, 0.1409609466791153, 0.03976983949542...
60
60
['Baolin Peng', 'Kaisheng Yao']
1506.00195v1
Recurrent Neural Networks (RNNs) have become increasingly popular for the task of language understanding. In this task, a semantic tagger is deployed to associate a semantic label to each word in an input sequence. The success of RNN may be attributed to its ability to memorize long-term dependence that relates the cur...
Recurrent Neural Networks with External Memory for Language Understanding
2,015
http://arxiv.org/pdf/1506.00195v1
Title Recurrent Neural Networks External Memory Language Understanding Summary Recurrent Neural Networks RNNs become increasingly popular task language understanding task semantic tagger deployed associate semantic label word input sequence success RNN may attributed ability memorize longterm dependence relates current...
[0.005397650878876448, -0.028821926563978195, -0.004661565646529198, 0.03165142610669136, -0.013744932599365711, -0.011485778726637363, -0.03155529126524925, -0.01816035993397236, -0.0040605636313557625, -0.04181218147277832, -0.017562074586749077, -0.01850859820842743, 0.015784334391355515, 0.018213815987110138, 0.000...
61
61
['Alessandro Sordoni', 'Michel Galley', 'Michael Auli', 'Chris Brockett', 'Yangfeng Ji', 'Margaret Mitchell', 'Jian-Yun Nie', 'Jianfeng Gao', 'Bill Dolan']
1506.06714v1
We present a novel response generation system that can be trained end to end on large quantities of unstructured Twitter conversations. A neural network architecture is used to address sparsity issues that arise when integrating contextual information into classic statistical models, allowing the system to take into ac...
A Neural Network Approach to Context-Sensitive Generation of Conversational Responses
2,015
http://arxiv.org/pdf/1506.06714v1
Title Neural Network Approach ContextSensitive Generation Conversational Responses Summary present novel response generation system trained end end large quantity unstructured Twitter conversation neural network architecture used address sparsity issue arise integrating contextual information classic statistical model ...
[0.06651893258094788, 0.02844914048910141, -0.015687178820371628, 0.01697389967739582, -0.011209608055651188, -0.00512953195720911, 0.022652603685855865, -0.0066788471303880215, 0.005032610148191452, -0.025628216564655304, 0.0077673643827438354, -0.02901383303105831, -0.008281727321445942, 0.06537818908691406, 0.010971...
62
62
['Ryan Lowe', 'Nissan Pow', 'Iulian Serban', 'Joelle Pineau']
1506.08909v3
This paper introduces the Ubuntu Dialogue Corpus, a dataset containing almost 1 million multi-turn dialogues, with a total of over 7 million utterances and 100 million words. This provides a unique resource for research into building dialogue managers based on neural language models that can make use of large amounts o...
The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems
2,015
http://arxiv.org/pdf/1506.08909v3
Title Ubuntu Dialogue Corpus Large Dataset Research Unstructured MultiTurn Dialogue Systems Summary paper introduces Ubuntu Dialogue Corpus dataset containing almost 1 million multiturn dialogue total 7 million utterance 100 million word provides unique resource research building dialogue manager based neural language ...
[0.05112911015748978, 0.06905069947242737, -0.0019867701921612024, 0.07173269987106323, -0.022180061787366867, 0.017659947276115417, -0.006270970683544874, -0.006695645395666361, 0.028420697897672653, -0.04165142402052879, -0.019953597337007523, -0.0066855731420218945, -0.016433417797088623, 0.053304798901081085, 0.006...
63
63
['Iulian V. Serban', 'Alessandro Sordoni', 'Yoshua Bengio', 'Aaron Courville', 'Joelle Pineau']
1507.04808v3
We investigate the task of building open domain, conversational dialogue systems based on large dialogue corpora using generative models. Generative models produce system responses that are autonomously generated word-by-word, opening up the possibility for realistic, flexible interactions. In support of this goal, we ...
Building End-To-End Dialogue Systems Using Generative Hierarchical Neural Network Models
2,015
http://arxiv.org/pdf/1507.04808v3
Title Building EndToEnd Dialogue Systems Using Generative Hierarchical Neural Network Models Summary investigate task building open domain conversational dialogue system based large dialogue corpus using generative model Generative model produce system response autonomously generated wordbyword opening possibility real...
[0.07504977285861969, 0.0586656853556633, 0.00667676841840148, 0.04332181066274643, 0.0008132587536238134, 0.005327644292265177, -0.00212988443672657, -0.013627498410642147, 0.012270418927073479, -0.0374271459877491, -0.005141301546245813, -0.04003185033798218, -0.0022301755379885435, 0.07689303159713745, 0.00879512820...
64
64
['Dzmitry Bahdanau', 'Jan Chorowski', 'Dmitriy Serdyuk', 'Philemon Brakel', 'Yoshua Bengio']
1508.04395v2
Many of the current state-of-the-art Large Vocabulary Continuous Speech Recognition Systems (LVCSR) are hybrids of neural networks and Hidden Markov Models (HMMs). Most of these systems contain separate components that deal with the acoustic modelling, language modelling and sequence decoding. We investigate a more dir...
End-to-End Attention-based Large Vocabulary Speech Recognition
2,015
http://arxiv.org/pdf/1508.04395v2
Title EndtoEnd Attentionbased Large Vocabulary Speech Recognition Summary Many current stateoftheart Large Vocabulary Continuous Speech Recognition Systems LVCSR hybrid neural network Hidden Markov Models HMMs system contain separate component deal acoustic modelling language modelling sequence decoding investigate dir...
[0.0036636502481997013, 0.03559265658259392, 0.028615348041057587, 0.05915956571698189, -0.004340333864092827, -0.011123569682240486, -8.60570726217702e-05, 0.006707873195409775, -0.019050870090723038, -0.044991590082645416, -0.0426768995821476, -0.014255586080253124, 0.03938936069607735, 0.03283065930008888, 0.0070450...
65
65
['Baolin Peng', 'Zhengdong Lu', 'Hang Li', 'Kam-Fai Wong']
1508.05508v1
We propose Neural Reasoner, a framework for neural network-based reasoning over natural language sentences. Given a question, Neural Reasoner can infer over multiple supporting facts and find an answer to the question in specific forms. Neural Reasoner has 1) a specific interaction-pooling mechanism, allowing it to exa...
Towards Neural Network-based Reasoning
2,015
http://arxiv.org/pdf/1508.05508v1
Title Towards Neural Networkbased Reasoning Summary propose Neural Reasoner framework neural networkbased reasoning natural language sentence Given question Neural Reasoner infer multiple supporting fact find answer question specific form Neural Reasoner 1 specific interactionpooling mechanism allowing examine multiple...
[0.018060922622680664, 0.03844502195715904, 0.02135317027568817, 0.053976576775312424, -0.029643064364790916, -8.704445644980296e-05, 0.0524468719959259, 0.0009235565667040646, -0.00865957885980606, -0.01684771478176117, 0.04920750856399536, 0.012071599252521992, -0.020566226914525032, 0.027884943410754204, 0.014114674...
66
66
['Hongyuan Mei', 'Mohit Bansal', 'Matthew R. Walter']
1509.00838v2
We propose an end-to-end, domain-independent neural encoder-aligner-decoder model for selective generation, i.e., the joint task of content selection and surface realization. Our model first encodes a full set of over-determined database event records via an LSTM-based recurrent neural network, then utilizes a novel co...
What to talk about and how? Selective Generation using LSTMs with Coarse-to-Fine Alignment
2,015
http://arxiv.org/pdf/1509.00838v2
Title talk Selective Generation using LSTMs CoarsetoFine Alignment Summary propose endtoend domainindependent neural encoderalignerdecoder model selective generation ie joint task content selection surface realization model first encodes full set overdetermined database event record via LSTMbased recurrent neural netwo...
[-0.0010701733408495784, 0.020024528726935387, -0.007494167424738407, 0.034683242440223694, 0.010953160002827644, -0.006938681937754154, 0.010699848644435406, -0.01741715520620346, -0.04157884791493416, -0.014847035519778728, 0.003465132787823677, 0.012246784754097462, 0.02054784819483757, 0.0906740054488182, 0.0112764...
67
67
['Tim Rocktäschel', 'Edward Grefenstette', 'Karl Moritz Hermann', 'Tomáš Kočiský', 'Phil Blunsom']
1509.06664v4
While most approaches to automatically recognizing entailment relations have used classifiers employing hand engineered features derived from complex natural language processing pipelines, in practice their performance has been only slightly better than bag-of-word pair classifiers using only lexical similarity. The on...
Reasoning about Entailment with Neural Attention
2,015
http://arxiv.org/pdf/1509.06664v4
Title Reasoning Entailment Neural Attention Summary approach automatically recognizing entailment relation used classifier employing hand engineered feature derived complex natural language processing pipeline practice performance slightly better bagofword pair classifier using lexical similarity attempt far build endt...
[0.040562331676483154, 0.017046701163053513, -0.009960235096514225, 0.0880603939294815, -0.037016768008470535, 0.01922035776078701, -0.026768028736114502, -0.005412455648183823, 0.04625948518514633, -0.03011850267648697, 0.022725339978933334, 0.046445656567811966, 0.012567591853439808, 0.03314181789755821, 0.0241728182...
68
68
['Yu Zhang', 'Guoguo Chen', 'Dong Yu', 'Kaisheng Yao', 'Sanjeev Khudanpur', 'James Glass']
1510.08983v2
In this paper, we extend the deep long short-term memory (DLSTM) recurrent neural networks by introducing gated direct connections between memory cells in adjacent layers. These direct links, called highway connections, enable unimpeded information flow across different layers and thus alleviate the gradient vanishing ...
Highway Long Short-Term Memory RNNs for Distant Speech Recognition
2,015
http://arxiv.org/pdf/1510.08983v2
Title Highway Long ShortTerm Memory RNNs Distant Speech Recognition Summary paper extend deep long shortterm memory DLSTM recurrent neural network introducing gated direct connection memory cell adjacent layer direct link called highway connection enable unimpeded information flow across different layer thus alleviate ...
[0.002620002953335643, 0.03971334174275398, 0.01974472403526306, 0.05725536122918129, 0.004510858561843634, -0.006223713047802448, 0.057537518441677094, 0.012117910198867321, 0.02173352614045143, -0.03259125351905823, -0.044787194579839706, -0.05307859182357788, 0.03997835889458656, 0.030911626294255257, 0.005527654662...
69
69
['Pengcheng Yin', 'Zhengdong Lu', 'Hang Li', 'Ben Kao']
1512.00965v2
We proposed Neural Enquirer as a neural network architecture to execute a natural language (NL) query on a knowledge-base (KB) for answers. Basically, Neural Enquirer finds the distributed representation of a query and then executes it on knowledge-base tables to obtain the answer as one of the values in the tables. Un...
Neural Enquirer: Learning to Query Tables with Natural Language
2,015
http://arxiv.org/pdf/1512.00965v2
Title Neural Enquirer Learning Query Tables Natural Language Summary proposed Neural Enquirer neural network architecture execute natural language NL query knowledgebase KB answer Basically Neural Enquirer find distributed representation query executes knowledgebase table obtain answer one value table Unlike similar ef...
[0.050602201372385025, 0.07603490352630615, -0.004970546346157789, 0.042139116674661636, -0.017126917839050293, -0.008230132050812244, -0.019356362521648407, 0.009580429643392563, 0.008824012242257595, -0.022665787488222122, -0.005566886160522699, 0.046493031084537506, -0.044214747846126556, 0.04636704921722412, 0.0340...
70
70
['Petr Baudiš', 'Jan Pichl', 'Tomáš Vyskočil', 'Jan Šedivý']
1603.06127v4
We review the task of Sentence Pair Scoring, popular in the literature in various forms - viewed as Answer Sentence Selection, Semantic Text Scoring, Next Utterance Ranking, Recognizing Textual Entailment, Paraphrasing or e.g. a component of Memory Networks. We argue that all such tasks are similar from the model per...
Sentence Pair Scoring: Towards Unified Framework for Text Comprehension
2,016
http://arxiv.org/pdf/1603.06127v4
Title Sentence Pair Scoring Towards Unified Framework Text Comprehension Summary review task Sentence Pair Scoring popular literature various form viewed Answer Sentence Selection Semantic Text Scoring Next Utterance Ranking Recognizing Textual Entailment Paraphrasing eg component Memory Networks argue task similar mod...
[0.04160737618803978, 0.04916667565703392, -0.006056435871869326, 0.07171273976564407, -0.0365428701043129, 0.023142997175455093, -0.002609814051538706, -0.029520545154809952, -0.012600592337548733, -0.03214128315448761, -0.02376331388950348, -0.011796790175139904, 0.02304866723716259, 0.028991086408495903, -0.01025020...
71
71
['Jiatao Gu', 'Zhengdong Lu', 'Hang Li', 'Victor O. K. Li']
1603.06393v3
We address an important problem in sequence-to-sequence (Seq2Seq) learning referred to as copying, in which certain segments in the input sequence are selectively replicated in the output sequence. A similar phenomenon is observable in human language communication. For example, humans tend to repeat entity names or eve...
Incorporating Copying Mechanism in Sequence-to-Sequence Learning
2,016
http://arxiv.org/pdf/1603.06393v3
Title Incorporating Copying Mechanism SequencetoSequence Learning Summary address important problem sequencetosequence Seq2Seq learning referred copying certain segment input sequence selectively replicated output sequence similar phenomenon observable human language communication example human tend repeat entity name ...
[0.0387384369969368, 0.08290021121501923, 0.001752879354171455, 0.03323785215616226, -0.050581447780132294, 0.007113951724022627, -0.007450574543327093, 0.0347331203520298, -0.0663609504699707, -0.0353577695786953, 0.015634460374712944, -0.0024770135059952736, 0.017113493755459785, 0.0466763935983181, 0.040870621800422...
72
72
['Iulian Vlad Serban', 'Alberto García-Durán', 'Caglar Gulcehre', 'Sungjin Ahn', 'Sarath Chandar', 'Aaron Courville', 'Yoshua Bengio']
1603.06807v2
Over the past decade, large-scale supervised learning corpora have enabled machine learning researchers to make substantial advances. However, to this date, there are no large-scale question-answer corpora available. In this paper we present the 30M Factoid Question-Answer Corpus, an enormous question answer pair corpu...
Generating Factoid Questions With Recurrent Neural Networks: The 30M Factoid Question-Answer Corpus
2,016
http://arxiv.org/pdf/1603.06807v2
Title Generating Factoid Questions Recurrent Neural Networks 30M Factoid QuestionAnswer Corpus Summary past decade largescale supervised learning corpus enabled machine learning researcher make substantial advance However date largescale questionanswer corpus available paper present 30M Factoid QuestionAnswer Corpus en...
[0.07906658947467804, 0.03347751125693321, -0.012156195007264614, 0.024543317034840584, -0.008402527309954166, 0.02561432495713234, 0.03566369786858559, 0.01811807043850422, -0.018982665613293648, -0.02428242564201355, 0.021681996062397957, -0.007761327549815178, -0.04212968796491623, 0.03326527774333954, 0.03074628487...
73
73
['Chia-Wei Liu', 'Ryan Lowe', 'Iulian V. Serban', 'Michael Noseworthy', 'Laurent Charlin', 'Joelle Pineau']
1603.08023v2
We investigate evaluation metrics for dialogue response generation systems where supervised labels, such as task completion, are not available. Recent works in response generation have adopted metrics from machine translation to compare a model's generated response to a single target response. We show that these metric...
How NOT To Evaluate Your Dialogue System: An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generation
2,016
http://arxiv.org/pdf/1603.08023v2
Title Evaluate Dialogue System Empirical Study Unsupervised Evaluation Metrics Dialogue Response Generation Summary investigate evaluation metric dialogue response generation system supervised label task completion available Recent work response generation adopted metric machine translation compare model generated resp...
[0.06705658882856369, 0.0024863320868462324, -0.010683062486350536, 0.026092534884810448, -0.011607684195041656, 0.007161720655858517, 0.020762262865900993, 0.0042042238637804985, 0.02620399184525013, -0.04520987719297409, -0.044234130531549454, -0.009241407737135887, -0.003777619218453765, 0.06811194121837616, -0.0059...
74
74
['Iulian Vlad Serban', 'Alessandro Sordoni', 'Ryan Lowe', 'Laurent Charlin', 'Joelle Pineau', 'Aaron Courville', 'Yoshua Bengio']
1605.06069v3
Sequential data often possesses a hierarchical structure with complex dependencies between subsequences, such as found between the utterances in a dialogue. In an effort to model this kind of generative process, we propose a neural network-based generative architecture, with latent stochastic variables that span a vari...
A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialogues
2,016
http://arxiv.org/pdf/1605.06069v3
Title Hierarchical Latent Variable EncoderDecoder Model Generating Dialogues Summary Sequential data often posse hierarchical structure complex dependency subsequence found utterance dialogue effort model kind generative process propose neural networkbased generative architecture latent stochastic variable span variabl...
[0.06646768748760223, 0.06293745338916779, -0.007975352928042412, 0.024971112608909607, 0.008771158754825592, -0.007252962794154882, 0.003914270084351301, -0.025218840688467026, -0.025848684832453728, -0.03928428888320923, 0.016512366011738777, -0.02540435455739498, 0.01444464735686779, 0.0966857448220253, 0.0244050584...
75
75
['Dirk Weissenborn']
1606.03864v2
Many important NLP problems can be posed as dual-sequence or sequence-to-sequence modeling tasks. Recent advances in building end-to-end neural architectures have been highly successful in solving such tasks. In this work we propose a new architecture for dual-sequence modeling that is based on associative memory. We d...
Neural Associative Memory for Dual-Sequence Modeling
2,016
http://arxiv.org/pdf/1606.03864v2
Title Neural Associative Memory DualSequence Modeling Summary Many important NLP problem posed dualsequence sequencetosequence modeling task Recent advance building endtoend neural architecture highly successful solving task work propose new architecture dualsequence modeling based associative memory derive AMRNNs recu...
[0.05189606919884682, 0.049359217286109924, -0.03858322277665138, 0.03595011308789253, -0.026050789281725883, 0.0005870053428225219, -0.008557391352951527, -0.03977277874946594, 0.00774132227525115, -0.047866933047771454, 0.034330904483795166, -0.07000040262937546, -0.007414250168949366, 0.019366860389709473, 0.0148836...
76
76
['Marc Dymetman', 'Chunyang Xiao']
1607.02467v2
We introduce LL-RNNs (Log-Linear RNNs), an extension of Recurrent Neural Networks that replaces the softmax output layer by a log-linear output layer, of which the softmax is a special case. This conceptually simple move has two main advantages. First, it allows the learner to combat training data sparsity by allowing ...
Log-Linear RNNs: Towards Recurrent Neural Networks with Flexible Prior Knowledge
2,016
http://arxiv.org/pdf/1607.02467v2
Title LogLinear RNNs Towards Recurrent Neural Networks Flexible Prior Knowledge Summary introduce LLRNNs LogLinear RNNs extension Recurrent Neural Networks replaces softmax output layer loglinear output layer softmax special case conceptually simple move two main advantage First allows learner combat training data spar...
[0.018478121608495712, 0.04983919858932495, 0.0189732126891613, 0.026548797264695168, -0.03243403509259224, -0.004975564777851105, 0.010365774855017662, -0.01698579266667366, -0.04211164265871048, -0.02693321369588375, -0.012811820022761822, -0.04480641335248947, 0.027834083884954453, 0.05144667997956276, 0.04944017156...
77
77
['Ondrej Bajgar', 'Rudolf Kadlec', 'Jan Kleindienst']
1610.00956v1
There is a practically unlimited amount of natural language data available. Still, recent work in text comprehension has focused on datasets which are small relative to current computing possibilities. This article is making a case for the community to move to larger data and as a step in that direction it is proposing...
Embracing data abundance: BookTest Dataset for Reading Comprehension
2,016
http://arxiv.org/pdf/1610.00956v1
Title Embracing data abundance BookTest Dataset Reading Comprehension Summary practically unlimited amount natural language data available Still recent work text comprehension focused datasets small relative current computing possibility article making case community move larger data step direction proposing BookTest n...
[0.025445127859711647, 0.03385183960199356, -0.012874915264546871, 0.03594973310828209, -0.02379055880010128, 0.024186957627534866, 0.03685464709997177, -0.0361676961183548, -0.015774870291352272, -0.04073069989681244, -0.010656061582267284, -0.012908689677715302, 0.03424857184290886, 0.0068349954672157764, 0.015691727...
78
78
['James Bradbury', 'Stephen Merity', 'Caiming Xiong', 'Richard Socher']
1611.01576v2
Recurrent neural networks are a powerful tool for modeling sequential data, but the dependence of each timestep's computation on the previous timestep's output limits parallelism and makes RNNs unwieldy for very long sequences. We introduce quasi-recurrent neural networks (QRNNs), an approach to neural sequence modelin...
Quasi-Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1611.01576v2
Title QuasiRecurrent Neural Networks Summary Recurrent neural network powerful tool modeling sequential data dependence timesteps computation previous timesteps output limit parallelism make RNNs unwieldy long sequence introduce quasirecurrent neural network QRNNs approach neural sequence modeling alternate convolution...
[0.015634862706065178, 0.07041216641664505, -0.012827531434595585, 0.017810285091400146, -0.03620230033993721, 0.011213905178010464, 0.013429611921310425, 0.008376345969736576, -0.011478225700557232, -0.019606759771704674, 0.009776151739060879, -0.02758096717298031, 0.022093847393989563, 0.043517790734767914, -0.013645...
79
79
['Jakob N. Foerster', 'Justin Gilmer', 'Jan Chorowski', 'Jascha Sohl-Dickstein', 'David Sussillo']
1611.09434v2
There exist many problem domains where the interpretability of neural network models is essential for deployment. Here we introduce a recurrent architecture composed of input-switched affine transformations - in other words an RNN without any explicit nonlinearities, but with input-dependent recurrent weights. This sim...
Input Switched Affine Networks: An RNN Architecture Designed for Interpretability
2,016
http://arxiv.org/pdf/1611.09434v2
Title Input Switched Affine Networks RNN Architecture Designed Interpretability Summary exist many problem domain interpretability neural network model essential deployment introduce recurrent architecture composed inputswitched affine transformation word RNN without explicit nonlinearities inputdependent recurrent wei...
[0.010674873366951942, 0.018224945291876793, -0.038852620869874954, 0.05739248916506767, -0.021765777841210365, -0.01853686012327671, 0.034835197031497955, 0.0044465819373726845, -0.06506939232349396, -0.02134864404797554, -0.030794117599725723, -0.0017826718976721168, 0.0074800411239266396, 0.0892687439918518, 0.02445...
80
80
['Michał Daniluk', 'Tim Rocktäschel', 'Johannes Welbl', 'Sebastian Riedel']
1702.04521v1
Neural language models predict the next token using a latent representation of the immediate token history. Recently, various methods for augmenting neural language models with an attention mechanism over a differentiable memory have been proposed. For predicting the next token, these models query information from a me...
Frustratingly Short Attention Spans in Neural Language Modeling
2,017
http://arxiv.org/pdf/1702.04521v1
Title Frustratingly Short Attention Spans Neural Language Modeling Summary Neural language model predict next token using latent representation immediate token history Recently various method augmenting neural language model attention mechanism differentiable memory proposed predicting next token model query informatio...
[0.04816904664039612, 0.05674388259649277, -0.0201411135494709, 0.01817266270518303, -0.007094713859260082, -0.011209775693714619, 0.017209338024258614, -0.02115270122885704, -0.04461260139942169, -0.0364251472055912, -0.006139329168945551, -0.03879533335566521, 0.01870228722691536, 0.05745745822787285, 0.0486161746084...
81
81
['Zhouhan Lin', 'Minwei Feng', 'Cicero Nogueira dos Santos', 'Mo Yu', 'Bing Xiang', 'Bowen Zhou', 'Yoshua Bengio']
1703.03130v1
This paper proposes a new model for extracting an interpretable sentence embedding by introducing self-attention. Instead of using a vector, we use a 2-D matrix to represent the embedding, with each row of the matrix attending on a different part of the sentence. We also propose a self-attention mechanism and a special...
A Structured Self-attentive Sentence Embedding
2,017
http://arxiv.org/pdf/1703.03130v1
Title Structured Selfattentive Sentence Embedding Summary paper proposes new model extracting interpretable sentence embedding introducing selfattention Instead using vector use 2D matrix represent embedding row matrix attending different part sentence also propose selfattention mechanism special regularization term mo...
[0.025598371401429176, 0.014826016500592232, -0.0061867861077189445, 0.09527096152305603, -0.023132028058171272, 0.004297102335840464, -0.009861559607088566, -0.03200750797986984, 0.048720844089984894, -0.04509393125772476, -0.01509841252118349, 0.065526582300663, -0.048346150666475296, 0.040342625230550766, -0.0106384...
82
82
['Samuel Rönnqvist', 'Niko Schenk', 'Christian Chiarcos']
1704.08092v1
We introduce an attention-based Bi-LSTM for Chinese implicit discourse relations and demonstrate that modeling argument pairs as a joint sequence can outperform word order-agnostic approaches. Our model benefits from a partial sampling scheme and is conceptually simple, yet achieves state-of-the-art performance on the ...
A Recurrent Neural Model with Attention for the Recognition of Chinese Implicit Discourse Relations
2,017
http://arxiv.org/pdf/1704.08092v1
Title Recurrent Neural Model Attention Recognition Chinese Implicit Discourse Relations Summary introduce attentionbased BiLSTM Chinese implicit discourse relation demonstrate modeling argument pair joint sequence outperform word orderagnostic approach model benefit partial sampling scheme conceptually simple yet achie...
[0.0388648621737957, 0.015349977649748325, -0.009019549936056137, 0.08634036034345627, -0.006240590941160917, -0.01864231750369072, 0.0005546743050217628, -0.009385770186781883, 0.009124504402279854, -0.07483068853616714, -0.003993363585323095, -0.02777963876724243, 0.02795795537531376, 0.013620425947010517, 0.00709872...
83
83
['Lara J. Martin', 'Prithviraj Ammanabrolu', 'Xinyu Wang', 'William Hancock', 'Shruti Singh', 'Brent Harrison', 'Mark O. Riedl']
1706.01331v3
Automated story generation is the problem of automatically selecting a sequence of events, actions, or words that can be told as a story. We seek to develop a system that can generate stories by learning everything it needs to know from textual story corpora. To date, recurrent neural networks that learn language model...
Event Representations for Automated Story Generation with Deep Neural Nets
2,017
http://arxiv.org/pdf/1706.01331v3
Title Event Representations Automated Story Generation Deep Neural Nets Summary Automated story generation problem automatically selecting sequence event action word told story seek develop system generate story learning everything need know textual story corpus date recurrent neural network learn language model charac...
[0.024326177313923836, 0.034451182931661606, -0.02322245202958584, 0.05542153865098953, -0.05120529234409332, -0.0012775624636560678, 0.022607892751693726, -0.012914764694869518, 0.0013930521672591567, -0.039871297776699066, 0.06197723001241684, 0.01201376412063837, 0.015386915765702724, 0.13734115660190582, -0.0196597...
84
84
['Tong Wang', 'Xingdi Yuan', 'Adam Trischler']
1706.01450v1
We propose a generative machine comprehension model that learns jointly to ask and answer questions based on documents. The proposed model uses a sequence-to-sequence framework that encodes the document and generates a question (answer) given an answer (question). Significant improvement in model performance is observe...
A Joint Model for Question Answering and Question Generation
2,017
http://arxiv.org/pdf/1706.01450v1
Title Joint Model Question Answering Question Generation Summary propose generative machine comprehension model learns jointly ask answer question based document proposed model us sequencetosequence framework encodes document generates question answer given answer question Significant improvement model performance obse...
[0.07029268145561218, 0.05171780288219452, -0.020893651992082596, 0.01403932273387909, -0.014812195673584938, 0.0352637879550457, 0.012427830137312412, -0.00030363400583155453, 0.006699640769511461, -0.0045855785720050335, 0.004981146659702063, -0.005575785879045725, -0.035014454275369644, 0.04250645264983177, 0.010775...
85
85
['Wei Wen', 'Yuxiong He', 'Samyam Rajbhandari', 'Minjia Zhang', 'Wenhan Wang', 'Fang Liu', 'Bin Hu', 'Yiran Chen', 'Hai Li']
1709.05027v7
Model compression is significant for the wide adoption of Recurrent Neural Networks (RNNs) in both user devices possessing limited resources and business clusters requiring quick responses to large-scale service requests. This work aims to learn structurally-sparse Long Short-Term Memory (LSTM) by reducing the sizes of...
Learning Intrinsic Sparse Structures within Long Short-Term Memory
2,017
http://arxiv.org/pdf/1709.05027v7
Title Learning Intrinsic Sparse Structures within Long ShortTerm Memory Summary Model compression significant wide adoption Recurrent Neural Networks RNNs user device possessing limited resource business cluster requiring quick response largescale service request work aim learn structurallysparse Long ShortTerm Memory ...
[0.0015198070323094726, 0.024757659062743187, 0.004226320888847113, 0.05461505427956581, 0.02564852125942707, -0.020287277176976204, 0.035923074930906296, 0.02002188377082348, -0.03183319792151451, -0.012838107533752918, -0.02065843902528286, -0.056550681591033936, 0.009196764789521694, 0.030382253229618073, 0.02733928...
86
86
['Huda Hakami', 'Danushka Bollegala', 'Hayashi Kohei']
1709.06673v2
Representing the semantic relations that exist between two given words (or entities) is an important first step in a wide-range of NLP applications such as analogical reasoning, knowledge base completion and relational information retrieval. A simple, yet surprisingly accurate method for representing a relation between...
Why PairDiff works? -- A Mathematical Analysis of Bilinear Relational Compositional Operators for Analogy Detection
2,017
http://arxiv.org/pdf/1709.06673v2
Title PairDiff work Mathematical Analysis Bilinear Relational Compositional Operators Analogy Detection Summary Representing semantic relation exist two given word entity important first step widerange NLP application analogical reasoning knowledge base completion relational information retrieval simple yet surprisingl...
[0.005939915776252747, 0.05347522720694542, -0.010577460750937462, 0.0691685602068901, -0.0343889556825161, 0.016756443306803703, 0.007281046360731125, 0.022173646837472916, 0.004108861088752747, -0.05177553743124008, -0.011418481357395649, -0.010816593654453754, -0.019512318074703217, -0.028440771624445915, 0.00320503...
87
87
['Zhengdong Lu', 'Haotian Cui', 'Xianggen Liu', 'Yukun Yan', 'Daqi Zheng']
1709.08853v4
We propose Object-oriented Neural Programming (OONP), a framework for semantically parsing documents in specific domains. Basically, OONP reads a document and parses it into a predesigned object-oriented data structure (referred to as ontology in this paper) that reflects the domain-specific semantics of the document. ...
Object-oriented Neural Programming (OONP) for Document Understanding
2,017
http://arxiv.org/pdf/1709.08853v4
Title Objectoriented Neural Programming OONP Document Understanding Summary propose Objectoriented Neural Programming OONP framework semantically parsing document specific domain Basically OONP read document par predesigned objectoriented data structure referred ontology paper reflects domainspecific semantics document...
[0.028467047959566116, 0.03453904017806053, 0.028609363362193108, 0.026870351284742355, -0.020189322531223297, 0.012441849336028099, -0.004567192867398262, -0.006413863971829414, -0.04142891615629196, -0.08128751069307327, -0.0010759899159893394, 0.06674836575984955, -0.017097996547818184, 0.07013848423957825, -0.05891...
88
88
['Bin Bi', 'Hao Ma']
1709.10204v2
This paper proposes a novel neural machine reading model for open-domain question answering at scale. Existing machine comprehension models typically assume that a short piece of relevant text containing answers is already identified and given to the models, from which the models are designed to extract answers. This a...
A Neural Comprehensive Ranker (NCR) for Open-Domain Question Answering
2,017
http://arxiv.org/pdf/1709.10204v2
Title Neural Comprehensive Ranker NCR OpenDomain Question Answering Summary paper proposes novel neural machine reading model opendomain question answering scale Existing machine comprehension model typically assume short piece relevant text containing answer already identified given model model designed extract answer...
[0.05983716621994972, 0.03970317915081978, 0.001613727188669145, 0.03362972289323807, -0.022936943918466568, 0.048818159848451614, 0.005240192171186209, -0.01142068300396204, 0.03304155170917511, -0.0009634695597924292, 0.01396422740072012, -0.021603025496006012, -0.039359331130981445, 0.040468744933605194, -0.00762926...
89
89
['Mirco Ravanelli', 'Philemon Brakel', 'Maurizio Omologo', 'Yoshua Bengio']
1710.00641v1
Speech recognition is largely taking advantage of deep learning, showing that substantial benefits can be obtained by modern Recurrent Neural Networks (RNNs). The most popular RNNs are Long Short-Term Memory (LSTMs), which typically reach state-of-the-art performance in many tasks thanks to their ability to learn long-...
Improving speech recognition by revising gated recurrent units
2,017
http://arxiv.org/pdf/1710.00641v1
Title Improving speech recognition revising gated recurrent unit Summary Speech recognition largely taking advantage deep learning showing substantial benefit obtained modern Recurrent Neural Networks RNNs popular RNNs Long ShortTerm Memory LSTMs typically reach stateoftheart performance many task thanks ability learn ...
[-0.011323253624141216, 0.04274016618728638, 0.02161041833460331, 0.06906063109636307, -0.012762226164340973, -0.02150745689868927, 0.054751697927713394, 0.015127060003578663, -0.014180455356836319, -0.04825058951973915, -0.04682881385087967, -0.04433776065707207, 0.063630610704422, 0.07375448197126389, 0.0187855027616...
90
90
['Baolin Peng', 'Xiujun Li', 'Jianfeng Gao', 'Jingjing Liu', 'Kam-Fai Wong']
1801.06176v1
Training a task-completion dialogue agent with real users via reinforcement learning (RL) could be prohibitively expensive, because it requires many interactions with users. One alternative is to resort to a user simulator, while the discrepancy of between simulated and real users makes the learned policy unreliable in...
Integrating planning for task-completion dialogue policy learning
2,018
http://arxiv.org/pdf/1801.06176v1
Title Integrating planning taskcompletion dialogue policy learning Summary Training taskcompletion dialogue agent real user via reinforcement learning RL could prohibitively expensive requires many interaction user One alternative resort user simulator discrepancy simulated real user make learned policy unreliable prac...
[0.052217062562704086, 0.045885395258665085, -0.004483434837311506, -0.023136097937822342, -0.023510579019784927, 0.009580758400261402, 0.003991948440670967, -0.031473308801651, -0.006071672774851322, -0.007947076112031937, -0.021096833050251007, 0.024964341893792152, -0.025524942204356194, 0.07312264293432236, -0.0373...
91
91
['Andrew L. Maas', 'Peng Qi', 'Ziang Xie', 'Awni Y. Hannun', 'Christopher T. Lengerich', 'Daniel Jurafsky', 'Andrew Y. Ng']
1406.7806v2
Deep neural networks (DNNs) are now a central component of nearly all state-of-the-art speech recognition systems. Building neural network acoustic models requires several design decisions including network architecture, size, and training loss function. This paper offers an empirical investigation on which aspects of ...
Building DNN Acoustic Models for Large Vocabulary Speech Recognition
2,014
http://arxiv.org/pdf/1406.7806v2
Title Building DNN Acoustic Models Large Vocabulary Speech Recognition Summary Deep neural network DNNs central component nearly stateoftheart speech recognition system Building neural network acoustic model requires several design decision including network architecture size training loss function paper offer empirica...
[0.008880573324859142, 0.018985655158758163, 0.01436031237244606, 0.03560292720794678, -0.003236984834074974, -0.026291022077202797, 0.05768342316150665, -0.005553683266043663, -0.011256711557507515, -0.022665170952677727, -0.0633264109492302, 0.002275430131703615, 0.05088605359196663, 0.013685393147170544, -0.00902808...
92
92
['William Chan', 'Ian Lane']
1504.01482v1
We present a novel deep Recurrent Neural Network (RNN) model for acoustic modelling in Automatic Speech Recognition (ASR). We term our contribution as a TC-DNN-BLSTM-DNN model, the model combines a Deep Neural Network (DNN) with Time Convolution (TC), followed by a Bidirectional Long Short-Term Memory (BLSTM), and a fi...
Deep Recurrent Neural Networks for Acoustic Modelling
2,015
http://arxiv.org/pdf/1504.01482v1
Title Deep Recurrent Neural Networks Acoustic Modelling Summary present novel deep Recurrent Neural Network RNN model acoustic modelling Automatic Speech Recognition ASR term contribution TCDNNBLSTMDNN model model combine Deep Neural Network DNN Time Convolution TC followed Bidirectional Long ShortTerm Memory BLSTM fin...
[-0.023029625415802002, -0.004876798950135708, 0.007106386125087738, 0.05048801377415657, -0.020864373072981834, -0.02969019114971161, 0.021685227751731873, -0.054400306195020676, -0.07343190908432007, -0.0060390206053853035, -0.02053234726190567, -0.012641951441764832, 0.05040573328733444, 0.03846380114555359, 0.01577...
93
93
['David Krueger', 'Roland Memisevic']
1511.08400v7
We stabilize the activations of Recurrent Neural Networks (RNNs) by penalizing the squared distance between successive hidden states' norms. This penalty term is an effective regularizer for RNNs including LSTMs and IRNNs, improving performance on character-level language modeling and phoneme recognition, and outperf...
Regularizing RNNs by Stabilizing Activations
2,015
http://arxiv.org/pdf/1511.08400v7
Title Regularizing RNNs Stabilizing Activations Summary stabilize activation Recurrent Neural Networks RNNs penalizing squared distance successive hidden state norm penalty term effective regularizer RNNs including LSTMs IRNNs improving performance characterlevel language modeling phoneme recognition outperforming weig...
[0.0009324587881565094, 0.07279003411531448, 0.008771753869950771, 0.014386710710823536, 0.0023166430182754993, -0.029290569946169853, 0.03192298859357834, 0.040976859629154205, -0.02445213869214058, 0.016157979145646095, -0.02064850553870201, -0.058204080909490585, 0.01878456398844719, 0.007700189482420683, 0.00976213...
94
94
['Noam Shazeer', 'Azalia Mirhoseini', 'Krzysztof Maziarz', 'Andy Davis', 'Quoc Le', 'Geoffrey Hinton', 'Jeff Dean']
1701.06538v1
The capacity of a neural network to absorb information is limited by its number of parameters. Conditional computation, where parts of the network are active on a per-example basis, has been proposed in theory as a way of dramatically increasing model capacity without a proportional increase in computation. In practice...
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
2,017
http://arxiv.org/pdf/1701.06538v1
Title Outrageously Large Neural Networks SparselyGated MixtureofExperts Layer Summary capacity neural network absorb information limited number parameter Conditional computation part network active perexample basis proposed theory way dramatically increasing model capacity without proportional increase computation prac...
[0.036124326288700104, 0.03961411491036415, -0.00015774837811477482, 0.06119773909449577, 0.02255227603018284, 0.016653865575790405, 0.03068103827536106, 0.028619125485420227, -0.04319979250431061, -0.026061443611979485, -0.02470974624156952, -0.05041300132870674, 0.009488970972597599, 0.01390055101364851, 0.0352874696...
95
95
['Yacine Jernite', 'Samuel R. Bowman', 'David Sontag']
1705.00557v1
This work presents a novel objective function for the unsupervised training of neural network sentence encoders. It exploits signals from paragraph-level discourse coherence to train these models to understand text. Our objective is purely discriminative, allowing us to train models many times faster than was possible ...
Discourse-Based Objectives for Fast Unsupervised Sentence Representation Learning
2,017
http://arxiv.org/pdf/1705.00557v1
Title DiscourseBased Objectives Fast Unsupervised Sentence Representation Learning Summary work present novel objective function unsupervised training neural network sentence encoders exploit signal paragraphlevel discourse coherence train model understand text objective purely discriminative allowing u train model man...
[0.009210407733917236, 0.04873969778418541, 0.031147388741374016, 0.05972865968942642, -0.03127425163984299, -0.01853874884545803, -0.0032950961031019688, -0.03150421380996704, -0.008737357333302498, -0.07214866578578949, -0.005271330941468477, 0.027711158618330956, 0.01594187691807747, 0.009612095542252064, -0.0387347...
96
96
['Zhengyang Wang', 'Shuiwang Ji']
1705.06824v1
Visual question answering is a recently proposed artificial intelligence task that requires a deep understanding of both images and texts. In deep learning, images are typically modeled through convolutional neural networks, and texts are typically modeled through recurrent neural networks. While the requirement for mo...
Learning Convolutional Text Representations for Visual Question Answering
2,017
http://arxiv.org/pdf/1705.06824v1
Title Learning Convolutional Text Representations Visual Question Answering Summary Visual question answering recently proposed artificial intelligence task requires deep understanding image text deep learning image typically modeled convolutional neural network text typically modeled recurrent neural network requireme...
[0.07197307795286179, 0.032304778695106506, -0.005745700094848871, 0.06601636111736298, -0.030158162117004395, 0.01168147474527359, 0.0482843816280365, 0.03252594918012619, -0.0017894002376124263, -0.048014212399721146, -0.0009359444375149906, -0.011657709255814552, -0.019961515441536903, 0.07049494981765747, 0.0264007...
97
97
['Kyunghyun Cho', 'Bart van Merrienboer', 'Caglar Gulcehre', 'Dzmitry Bahdanau', 'Fethi Bougares', 'Holger Schwenk', 'Yoshua Bengio']
1406.1078v3
In this paper, we propose a novel neural network model called RNN Encoder-Decoder that consists of two recurrent neural networks (RNN). One RNN encodes a sequence of symbols into a fixed-length vector representation, and the other decodes the representation into another sequence of symbols. The encoder and decoder of t...
Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation
2,014
http://arxiv.org/pdf/1406.1078v3
Title Learning Phrase Representations using RNN EncoderDecoder Statistical Machine Translation Summary paper propose novel neural network model called RNN EncoderDecoder consists two recurrent neural network RNN One RNN encodes sequence symbol fixedlength vector representation decodes representation another sequence sy...
[0.04529965668916702, 0.026439426466822624, -0.013809925876557827, 0.051517702639102936, -0.06017933413386345, 0.012948517687618732, -0.02087404578924179, 0.005684196949005127, -0.05261906981468201, -0.06446237862110138, 0.006176763214170933, -0.00038310239324346185, 0.026355110108852386, 0.004713333677500486, -0.00273...
98
98
['Zhiyuan Tang', 'Dong Wang', 'Zhiyong Zhang']
1505.04630v5
Recurrent neural networks (RNNs), particularly long short-term memory (LSTM), have gained much attention in automatic speech recognition (ASR). Although some successful stories have been reported, training RNNs remains highly challenging, especially with limited training data. Recent research found that a well-trained ...
Recurrent Neural Network Training with Dark Knowledge Transfer
2,015
http://arxiv.org/pdf/1505.04630v5
Title Recurrent Neural Network Training Dark Knowledge Transfer Summary Recurrent neural network RNNs particularly long shortterm memory LSTM gained much attention automatic speech recognition ASR Although successful story reported training RNNs remains highly challenging especially limited training data Recent researc...
[0.01449339184910059, 0.026175323873758316, 0.0177297405898571, 0.04964710772037506, -0.008629859425127506, -0.005975619424134493, 0.03218740597367287, -0.02890104427933693, -0.05938320606946945, -0.03964444249868393, -0.058293748646974564, -0.00953069981187582, 0.027942024171352386, 0.006167636718600988, 0.01253994181...
99
99
['Haşim Sak', 'Andrew Senior', 'Françoise Beaufays']
1402.1128v1
Long Short-Term Memory (LSTM) is a recurrent neural network (RNN) architecture that has been designed to address the vanishing and exploding gradient problems of conventional RNNs. Unlike feedforward neural networks, RNNs have cyclic connections making them powerful for modeling sequences. They have been successfully u...
Long Short-Term Memory Based Recurrent Neural Network Architectures for Large Vocabulary Speech Recognition
2,014
http://arxiv.org/pdf/1402.1128v1
Title Long ShortTerm Memory Based Recurrent Neural Network Architectures Large Vocabulary Speech Recognition Summary Long ShortTerm Memory LSTM recurrent neural network RNN architecture designed address vanishing exploding gradient problem conventional RNNs Unlike feedforward neural network RNNs cyclic connection makin...
[0.0024354096967726946, -0.0008057195809669793, 0.014775389805436134, 0.06812814623117447, 0.011045847088098526, -0.021482767537236214, 0.03765282779932022, -0.0010265184100717306, -0.00853055901825428, -0.03915851190686226, -0.05398586392402649, -0.031463298946619034, 0.04310267046093941, 0.008393342606723309, 0.00355...