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
500
500
['Ji Young Lee', 'Franck Dernoncourt', 'Ozlem Uzuner', 'Peter Szolovits']
1610.09704v1
Patient notes contain a wealth of information of potentially great interest to medical investigators. However, to protect patients' privacy, Protected Health Information (PHI) must be removed from the patient notes before they can be legally released, a process known as patient note de-identification. The main objectiv...
Feature-Augmented Neural Networks for Patient Note De-identification
2,016
http://arxiv.org/pdf/1610.09704v1
Title FeatureAugmented Neural Networks Patient Note Deidentification Summary Patient note contain wealth information potentially great interest medical investigator However protect patient privacy Protected Health Information PHI must removed patient note legally released process known patient note deidentification mai...
[0.043702758848667145, 0.08156780898571014, -0.015921566635370255, -0.013011055067181587, -0.005612940061837435, 0.007986437529325485, 0.04966949298977852, 0.019367963075637817, 0.006645694375038147, 0.005578699987381697, 0.03567885980010033, -0.015435667708516121, 0.021586939692497253, 0.08970491588115692, 0.011171227...
501
501
['Kartik Audhkhasi', 'Bhuvana Ramabhadran', 'George Saon', 'Michael Picheny', 'David Nahamoo']
1703.07754v1
Recent work on end-to-end automatic speech recognition (ASR) has shown that the connectionist temporal classification (CTC) loss can be used to convert acoustics to phone or character sequences. Such systems are used with a dictionary and separately-trained Language Model (LM) to produce word sequences. However, they a...
Direct Acoustics-to-Word Models for English Conversational Speech Recognition
2,017
http://arxiv.org/pdf/1703.07754v1
Title Direct AcousticstoWord Models English Conversational Speech Recognition Summary Recent work endtoend automatic speech recognition ASR shown connectionist temporal classification CTC loss used convert acoustic phone character sequence system used dictionary separatelytrained Language Model LM produce word sequence...
[0.02484346553683281, 0.033995211124420166, 0.01255255751311779, 0.0641455352306366, -0.03553905338048935, -0.00043097714660689235, 0.03510015085339546, 0.015079144388437271, -0.02754133567214012, -0.04431154206395149, -0.0484241284430027, -0.020511770620942116, 0.047198012471199036, 0.029331816360354424, -0.0065767988...
502
502
['Oleksii Kuchaiev', 'Boris Ginsburg']
1703.10722v3
We present two simple ways of reducing the number of parameters and accelerating the training of large Long Short-Term Memory (LSTM) networks: the first one is "matrix factorization by design" of LSTM matrix into the product of two smaller matrices, and the second one is partitioning of LSTM matrix, its inputs and stat...
Factorization tricks for LSTM networks
2,017
http://arxiv.org/pdf/1703.10722v3
Title Factorization trick LSTM network Summary present two simple way reducing number parameter accelerating training large Long ShortTerm Memory LSTM network first one matrix factorization design LSTM matrix product two smaller matrix second one partitioning LSTM matrix input state independent group approach allow u t...
[-0.026453623548150063, 0.03137867525219917, 0.014112189412117004, 0.07220063358545303, 0.005589200649410486, 0.00911327637732029, 0.02947700396180153, 0.0013658213429152966, -0.03214815631508827, -0.04276707395911217, -0.046675652265548706, 0.0071130963042378426, 0.0020036331843584776, 0.03333934023976326, 0.021827960...
503
503
['Franck Dernoncourt', 'Ji Young Lee', 'Peter Szolovits']
1705.05487v1
Named-entity recognition (NER) aims at identifying entities of interest in a text. Artificial neural networks (ANNs) have recently been shown to outperform existing NER systems. However, ANNs remain challenging to use for non-expert users. In this paper, we present NeuroNER, an easy-to-use named-entity recognition tool...
NeuroNER: an easy-to-use program for named-entity recognition based on neural networks
2,017
http://arxiv.org/pdf/1705.05487v1
Title NeuroNER easytouse program namedentity recognition based neural network Summary Namedentity recognition NER aim identifying entity interest text Artificial neural network ANNs recently shown outperform existing NER system However ANNs remain challenging use nonexpert user paper present NeuroNER easytouse namedent...
[0.047105710953474045, 0.01603306084871292, 0.012159192003309727, 0.03783440217375755, 0.01361879613250494, 0.015137418173253536, 0.03210264444351196, 0.0026426741387695074, -0.007367223035544157, -0.03166850656270981, -0.0022698300890624523, 0.03770449012517929, 0.011607103981077671, 0.0037961245980113745, 0.023154465...
504
504
['Zhenisbek Assylbekov', 'Rustem Takhanov', 'Bagdat Myrzakhmetov', 'Jonathan N. Washington']
1707.06480v1
Syllabification does not seem to improve word-level RNN language modeling quality when compared to character-based segmentation. However, our best syllable-aware language model, achieving performance comparable to the competitive character-aware model, has 18%-33% fewer parameters and is trained 1.2-2.2 times faster.
Syllable-aware Neural Language Models: A Failure to Beat Character-aware Ones
2,017
http://arxiv.org/pdf/1707.06480v1
Title Syllableaware Neural Language Models Failure Beat Characteraware Ones Summary Syllabification seem improve wordlevel RNN language modeling quality compared characterbased segmentation However best syllableaware language model achieving performance comparable competitive characteraware model 1833 fewer parameter t...
[0.017384110018610954, 0.08129464089870453, 0.02031603455543518, 0.02979023940861225, -0.022303177043795586, -0.01334854494780302, 0.07097753137350082, 0.008717306889593601, -0.08133973926305771, -0.004194943234324455, -0.033547624945640564, -0.07666829973459244, 0.06709268689155579, -0.05620016157627106, 0.06638380140...
505
505
['Iñigo Casanueva', 'Paweł Budzianowski', 'Pei-Hao Su', 'Nikola Mrkšić', 'Tsung-Hsien Wen', 'Stefan Ultes', 'Lina Rojas-Barahona', 'Steve Young', 'Milica Gašić']
1711.11023v1
Dialogue assistants are rapidly becoming an indispensable daily aid. To avoid the significant effort needed to hand-craft the required dialogue flow, the Dialogue Management (DM) module can be cast as a continuous Markov Decision Process (MDP) and trained through Reinforcement Learning (RL). Several RL models have been...
A Benchmarking Environment for Reinforcement Learning Based Task Oriented Dialogue Management
2,017
http://arxiv.org/pdf/1711.11023v1
Title Benchmarking Environment Reinforcement Learning Based Task Oriented Dialogue Management Summary Dialogue assistant rapidly becoming indispensable daily aid avoid significant effort needed handcraft required dialogue flow Dialogue Management DM module cast continuous Markov Decision Process MDP trained Reinforceme...
[0.07061508297920227, 0.004467794205993414, -0.0014938859967514873, 0.006259709130972624, -0.043233588337898254, 0.005580393597483635, 0.024448717013001442, -0.03563292324542999, -0.012285783886909485, -0.01634962111711502, -0.03158413618803024, 0.0056253825314342976, -0.03055431693792343, 0.0707760900259018, -0.041852...
506
506
['Zhenisbek Assylbekov', 'Rustem Takhanov']
1802.08375v1
We propose several ways of reusing subword embeddings and other weights in subword-aware neural language models. The proposed techniques do not benefit a competitive character-aware model, but some of them improve the performance of syllable- and morpheme-aware models while showing significant reductions in model sizes...
Reusing Weights in Subword-aware Neural Language Models
2,018
http://arxiv.org/pdf/1802.08375v1
Title Reusing Weights Subwordaware Neural Language Models Summary propose several way reusing subword embeddings weight subwordaware neural language model proposed technique benefit competitive characteraware model improve performance syllable morphemeaware model showing significant reduction model size discover simple...
[0.0314902625977993, 0.06977414339780807, 0.009921452961862087, 0.05988321453332901, -0.03749609366059303, -0.014724734239280224, 0.026353269815444946, 0.018781455233693123, -0.03120853565633297, 0.010252118110656738, -0.02827058918774128, -0.05689524859189987, 0.03973189741373062, -0.0033906486351042986, 0.05414461717...
507
507
['Isabelle Augenstein', 'Sebastian Ruder', 'Anders Søgaard']
1802.09913v1
We combine multi-task learning and semi-supervised learning by inducing a joint embedding space between disparate label spaces and learning transfer functions between label embeddings, enabling us to jointly leverage unlabelled data and auxiliary, annotated datasets. We evaluate our approach on a variety of sequence cl...
Multi-task Learning of Pairwise Sequence Classification Tasks Over Disparate Label Spaces
2,018
http://arxiv.org/pdf/1802.09913v1
Title Multitask Learning Pairwise Sequence Classification Tasks Disparate Label Spaces Summary combine multitask learning semisupervised learning inducing joint embedding space disparate label space learning transfer function label embeddings enabling u jointly leverage unlabelled data auxiliary annotated datasets eval...
[-0.0072976648807525635, 0.029084190726280212, -0.015074536204338074, 0.01611342467367649, -0.002867706585675478, 0.03876494988799095, 0.0400647297501564, -0.03494870290160179, 0.029712270945310593, -0.06404870003461838, -0.040871720761060715, -0.019969899207353592, -0.03229004889726639, 0.047194626182317734, -0.001758...
508
508
['Caiming Xiong', 'Stephen Merity', 'Richard Socher']
1603.01417v1
Neural network architectures with memory and attention mechanisms exhibit certain reasoning capabilities required for question answering. One such architecture, the dynamic memory network (DMN), obtained high accuracy on a variety of language tasks. However, it was not shown whether the architecture achieves strong res...
Dynamic Memory Networks for Visual and Textual Question Answering
2,016
http://arxiv.org/pdf/1603.01417v1
Title Dynamic Memory Networks Visual Textual Question Answering Summary Neural network architecture memory attention mechanism exhibit certain reasoning capability required question answering One architecture dynamic memory network DMN obtained high accuracy variety language task However shown whether architecture achi...
[0.05473654344677925, -0.014790971763432026, -0.048675354570150375, 0.033531010150909424, 0.00014234644186217338, 0.024418624117970467, -0.011225576512515545, -0.001926077646203339, 0.0046080113388597965, -0.028277456760406494, 0.02012714184820652, -0.0230622086673975, -0.016158970072865486, 0.049191515892744064, 0.051...
509
509
['Fabio Carrara', 'Andrea Esuli', 'Tiziano Fagni', 'Fabrizio Falchi', 'Alejandro Moreo Fernández']
1606.07287v1
In this paper we tackle the problem of image search when the query is a short textual description of the image the user is looking for. We choose to implement the actual search process as a similarity search in a visual feature space, by learning to translate a textual query into a visual representation. Searching in t...
Picture It In Your Mind: Generating High Level Visual Representations From Textual Descriptions
2,016
http://arxiv.org/pdf/1606.07287v1
Title Picture Mind Generating High Level Visual Representations Textual Descriptions Summary paper tackle problem image search query short textual description image user looking choose implement actual search process similarity search visual feature space learning translate textual query visual representation Searching...
[0.03907022997736931, 0.039165105670690536, 0.004019522573798895, 0.06499049067497253, -0.020583219826221466, 0.005293701775372028, -0.0029849207494407892, 0.021723872050642967, -0.015071997418999672, -0.053929056972265244, -0.03275226429104805, 0.00982444453984499, -0.003892771666869521, 0.06876673549413681, 0.0166816...
510
510
['Marc Tanti', 'Albert Gatt', 'Kenneth P. Camilleri']
1703.09137v2
When a recurrent neural network language model is used for caption generation, the image information can be fed to the neural network either by directly incorporating it in the RNN -- conditioning the language model by `injecting' image features -- or in a layer following the RNN -- conditioning the language model by `...
Where to put the Image in an Image Caption Generator
2,017
http://arxiv.org/pdf/1703.09137v2
Title put Image Image Caption Generator Summary recurrent neural network language model used caption generation image information fed neural network either directly incorporating RNN conditioning language model injecting image feature layer following RNN conditioning language model merging image feature option attested...
[0.03209516778588295, 0.055982545018196106, -0.006379342172294855, 0.030374499037861824, -0.03485017269849777, -0.0006257386994548142, 0.035780780017375946, -0.0061155143193900585, -0.026076918467879295, -0.04781739413738251, -0.006454643327742815, -0.02906055562198162, 0.05437486618757248, 0.07614684104919434, 0.03335...
511
511
['Ilija Ilievski', 'Shuicheng Yan', 'Jiashi Feng']
1604.01485v1
Visual Question and Answering (VQA) problems are attracting increasing interest from multiple research disciplines. Solving VQA problems requires techniques from both computer vision for understanding the visual contents of a presented image or video, as well as the ones from natural language processing for understandi...
A Focused Dynamic Attention Model for Visual Question Answering
2,016
http://arxiv.org/pdf/1604.01485v1
Title Focused Dynamic Attention Model Visual Question Answering Summary Visual Question Answering VQA problem attracting increasing interest multiple research discipline Solving VQA problem requires technique computer vision understanding visual content presented image video well one natural language processing underst...
[0.06563566625118256, -0.00966651365160942, -0.027689313516020775, 0.04710132256150246, -0.010935337282717228, 0.02032623253762722, -0.0013326873304322362, 0.021615877747535706, 0.002249724231660366, -0.04811464250087738, 0.02061512880027294, -0.022769127041101456, -0.014294068329036236, 0.0656609833240509, 0.052788011...
512
512
['Remi Lebret', 'Pedro O. Pinheiro', 'Ronan Collobert']
1412.8419v3
Generating a novel textual description of an image is an interesting problem that connects computer vision and natural language processing. In this paper, we present a simple model that is able to generate descriptive sentences given a sample image. This model has a strong focus on the syntax of the descriptions. We tr...
Simple Image Description Generator via a Linear Phrase-Based Approach
2,014
http://arxiv.org/pdf/1412.8419v3
Title Simple Image Description Generator via Linear PhraseBased Approach Summary Generating novel textual description image interesting problem connects computer vision natural language processing paper present simple model able generate descriptive sentence given sample image model strong focus syntax description trai...
[0.04166076332330704, 0.03131520375609398, 0.03395695239305496, 0.07672151923179626, -0.032109953463077545, 0.0065893144346773624, 0.007731662131845951, 0.014008475467562675, -0.014690528623759747, -0.08661362528800964, 0.016673613339662552, 0.01575004868209362, 0.006975891999900341, 0.06346464902162552, 0.040673106908...
513
513
['Lin Ma', 'Zhengdong Lu', 'Lifeng Shang', 'Hang Li']
1504.06063v5
In this paper, we propose multimodal convolutional neural networks (m-CNNs) for matching image and sentence. Our m-CNN provides an end-to-end framework with convolutional architectures to exploit image representation, word composition, and the matching relations between the two modalities. More specifically, it consist...
Multimodal Convolutional Neural Networks for Matching Image and Sentence
2,015
http://arxiv.org/pdf/1504.06063v5
Title Multimodal Convolutional Neural Networks Matching Image Sentence Summary paper propose multimodal convolutional neural network mCNNs matching image sentence mCNN provides endtoend framework convolutional architecture exploit image representation word composition matching relation two modality specifically consist...
[0.04177653044462204, 0.06474702805280685, -0.0027690697461366653, 0.08906528353691101, -0.05987587571144104, -0.008205855265259743, 0.002720879390835762, 0.026942765340209007, -0.05217777565121651, -0.05570228770375252, -0.0026668068021535873, 0.007719388231635094, -0.03881571441888809, 0.04712129756808281, 0.00684999...
514
514
['Jacob Andreas', 'Marcus Rohrbach', 'Trevor Darrell', 'Dan Klein']
1601.01705v4
We describe a question answering model that applies to both images and structured knowledge bases. The model uses natural language strings to automatically assemble neural networks from a collection of composable modules. Parameters for these modules are learned jointly with network-assembly parameters via reinforcemen...
Learning to Compose Neural Networks for Question Answering
2,016
http://arxiv.org/pdf/1601.01705v4
Title Learning Compose Neural Networks Question Answering Summary describe question answering model applies image structured knowledge base model us natural language string automatically assemble neural network collection composable module Parameters module learned jointly networkassembly parameter via reinforcement le...
[0.0675688087940216, 0.008842003531754017, -0.012493891641497612, 0.025583632290363312, 3.815089803538285e-05, 0.02137421816587448, 0.0021996235009282827, 0.0015209135599434376, 0.00223616068251431, -0.03745744377374649, -0.006605775095522404, 0.005751275923103094, -0.02125655673444271, 0.06110196188092232, 0.015024741...
515
515
['Taehwan Kim', 'Weiran Wang', 'Hao Tang', 'Karen Livescu']
1602.04278v1
We study the problem of recognition of fingerspelled letter sequences in American Sign Language in a signer-independent setting. Fingerspelled sequences are both challenging and important to recognize, as they are used for many content words such as proper nouns and technical terms. Previous work has shown that it is p...
Signer-independent Fingerspelling Recognition with Deep Neural Network Adaptation
2,016
http://arxiv.org/pdf/1602.04278v1
Title Signerindependent Fingerspelling Recognition Deep Neural Network Adaptation Summary study problem recognition fingerspelled letter sequence American Sign Language signerindependent setting Fingerspelled sequence challenging important recognize used many content word proper noun technical term Previous work shown ...
[0.005663355812430382, 0.042473867535591125, 0.011422141455113888, 0.00483107753098011, -0.011012137867510319, 0.011493192985653877, 0.051666442304849625, -0.010203853249549866, 0.014243020676076412, -0.012097612023353577, -0.023107748478651047, -0.008776439353823662, 0.061435308307409286, 0.051392752677202225, 0.00125...
516
516
['Armand Vilalta', 'Dario Garcia-Gasulla', 'Ferran Parés', 'Eduard Ayguadé', 'Jesus Labarta', 'Ulises Cortés', 'Toyotaro Suzumura']
1707.09872v2
The current state-of-the-art for image annotation and image retrieval tasks is obtained through deep neural networks, which combine an image representation and a text representation into a shared embedding space. In this paper we evaluate the impact of using the Full-Network embedding in this setting, replacing the ori...
Full-Network Embedding in a Multimodal Embedding Pipeline
2,017
http://arxiv.org/pdf/1707.09872v2
Title FullNetwork Embedding Multimodal Embedding Pipeline Summary current stateoftheart image annotation image retrieval task obtained deep neural network combine image representation text representation shared embedding space paper evaluate impact using FullNetwork embedding setting replacing original image representa...
[0.012527558021247387, 0.01966680958867073, -0.009844220243394375, 0.037345778197050095, -0.009412907995283604, 0.007965855300426483, 0.04769733548164368, 0.0024188843090087175, 0.0004167724691797048, -0.0063164448365569115, -0.008858357556164265, -0.02805659919977188, -0.03822551667690277, 0.03703112527728081, 0.04509...
517
517
['Marc Tanti', 'Albert Gatt', 'Kenneth P. Camilleri']
1708.02043v2
In neural image captioning systems, a recurrent neural network (RNN) is typically viewed as the primary `generation' component. This view suggests that the image features should be `injected' into the RNN. This is in fact the dominant view in the literature. Alternatively, the RNN can instead be viewed as only encoding...
What is the Role of Recurrent Neural Networks (RNNs) in an Image Caption Generator?
2,017
http://arxiv.org/pdf/1708.02043v2
Title Role Recurrent Neural Networks RNNs Image Caption Generator Summary neural image captioning system recurrent neural network RNN typically viewed primary generation component view suggests image feature injected RNN fact dominant view literature Alternatively RNN instead viewed encoding previously generated word v...
[0.05952994152903557, 0.028533458709716797, 0.012885808013379574, 0.037569984793663025, -0.018395910039544106, 0.0034104178193956614, 0.02925185114145279, -0.025092288851737976, -0.025279764086008072, -0.05118748918175697, 0.006764248013496399, -0.03523680567741394, 0.025941558182239532, 0.06995707005262375, 0.02410381...
518
518
['Shiliang Zhang', 'Hui Jiang', 'Mingbin Xu', 'Junfeng Hou', 'Lirong Dai']
1505.01504v2
In this paper, we propose the new fixed-size ordinally-forgetting encoding (FOFE) method, which can almost uniquely encode any variable-length sequence of words into a fixed-size representation. FOFE can model the word order in a sequence using a simple ordinally-forgetting mechanism according to the positions of words...
A Fixed-Size Encoding Method for Variable-Length Sequences with its Application to Neural Network Language Models
2,015
http://arxiv.org/pdf/1505.01504v2
Title FixedSize Encoding Method VariableLength Sequences Application Neural Network Language Models Summary paper propose new fixedsize ordinallyforgetting encoding FOFE method almost uniquely encode variablelength sequence word fixedsize representation FOFE model word order sequence using simple ordinallyforgetting me...
[0.020751405507326126, 0.08787958323955536, -0.0008128008921630681, 0.01903299055993557, -0.02275308594107628, -0.0176555123180151, 0.008528110571205616, 0.015340976417064667, -0.04496227949857712, -0.02266302891075611, 0.03872222453355789, -0.06334277987480164, 0.0518856905400753, 0.02693445049226284, 0.04259736463427...
519
519
['Chris Dyer', 'Miguel Ballesteros', 'Wang Ling', 'Austin Matthews', 'Noah A. Smith']
1505.08075v1
We propose a technique for learning representations of parser states in transition-based dependency parsers. Our primary innovation is a new control structure for sequence-to-sequence neural networks---the stack LSTM. Like the conventional stack data structures used in transition-based parsing, elements can be pushed t...
Transition-Based Dependency Parsing with Stack Long Short-Term Memory
2,015
http://arxiv.org/pdf/1505.08075v1
Title TransitionBased Dependency Parsing Stack Long ShortTerm Memory Summary propose technique learning representation parser state transitionbased dependency parser primary innovation new control structure sequencetosequence neural networksthe stack LSTM Like conventional stack data structure used transitionbased pars...
[0.046386729925870895, 0.04305907338857651, -0.0023133051581680775, 0.036666374653577805, -0.03569711372256279, -0.025922877714037895, -0.02053554356098175, -0.004047680180519819, -0.011525302194058895, -0.0806916207075119, 0.03160813823342323, 0.01739772967994213, -0.00100998068228364, 0.08142969012260437, -0.02886826...
520
520
['Ehud Ben-Reuven', 'Jacob Goldberger']
1604.00317v1
In this study we address the problem of training a neuralnetwork for language identification using both labeled and unlabeled speech samples in the form of i-vectors. We propose a neural network architecture that can also handle out-of-set languages. We utilize a modified version of the recently proposed Ladder Network...
A Semisupervised Approach for Language Identification based on Ladder Networks
2,016
http://arxiv.org/pdf/1604.00317v1
Title Semisupervised Approach Language Identification based Ladder Networks Summary study address problem training neuralnetwork language identification using labeled unlabeled speech sample form ivectors propose neural network architecture also handle outofset language utilize modified version recently proposed Ladder...
[-0.007939611561596394, 0.06225118413567543, 0.018980644643306732, 0.03386964648962021, -0.001068435492925346, 0.006218391470611095, 0.03935648873448372, -0.029958978295326233, 0.032825157046318054, 0.0208926722407341, -0.0561082661151886, -0.012975585646927357, 0.03465845808386803, 0.012202487327158451, 0.013990128412...
521
521
['Awni Y. Hannun', 'Andrew L. Maas', 'Daniel Jurafsky', 'Andrew Y. Ng']
1408.2873v2
We present a method to perform first-pass large vocabulary continuous speech recognition using only a neural network and language model. Deep neural network acoustic models are now commonplace in HMM-based speech recognition systems, but building such systems is a complex, domain-specific task. Recent work demonstrated...
First-Pass Large Vocabulary Continuous Speech Recognition using Bi-Directional Recurrent DNNs
2,014
http://arxiv.org/pdf/1408.2873v2
Title FirstPass Large Vocabulary Continuous Speech Recognition using BiDirectional Recurrent DNNs Summary present method perform firstpass large vocabulary continuous speech recognition using neural network language model Deep neural network acoustic model commonplace HMMbased speech recognition system building system ...
[-0.03477975353598595, 0.03933965042233467, 0.013139639981091022, 0.0683111846446991, -0.027946796268224716, -0.005735600367188454, 0.019245104864239693, 0.008267185650765896, -0.013944591395556927, -0.0594002865254879, -0.0265803225338459, -0.041861556470394135, 0.01546946819871664, 0.056132446974515915, -0.0185661390...
522
522
['Jose Antonio Miñarro-Giménez', 'Oscar Marín-Alonso', 'Matthias Samwald']
1502.03682v1
BACKGROUND: The amount of biomedical literature is rapidly growing and it is becoming increasingly difficult to keep manually curated knowledge bases and ontologies up-to-date. In this study we applied the word2vec deep learning toolkit to medical corpora to test its potential for identifying relationships from unstruc...
Applying deep learning techniques on medical corpora from the World Wide Web: a prototypical system and evaluation
2,015
http://arxiv.org/pdf/1502.03682v1
Title Applying deep learning technique medical corpus World Wide Web prototypical system evaluation Summary BACKGROUND amount biomedical literature rapidly growing becoming increasingly difficult keep manually curated knowledge base ontology uptodate study applied word2vec deep learning toolkit medical corpus test pote...
[0.04419950023293495, 0.03953353315591812, 0.009901393204927444, -0.014953557401895523, -0.00039755282341502607, 0.03156756982207298, 0.005024713464081287, 0.053299590945243835, -0.008856860920786858, -0.023115988820791245, 0.029011059552431107, -0.04097963497042656, 0.029397210106253624, 0.05877178534865379, -0.009140...
523
523
['Mingxuan Wang', 'Zhengdong Lu', 'Hang Li', 'Qun Liu']
1503.02427v6
Many tasks in natural language processing, ranging from machine translation to question answering, can be reduced to the problem of matching two sentences or more generally two short texts. We propose a new approach to the problem, called Deep Match Tree (DeepMatch$_{tree}$), under a general setting. The approach consi...
Syntax-based Deep Matching of Short Texts
2,015
http://arxiv.org/pdf/1503.02427v6
Title Syntaxbased Deep Matching Short Texts Summary Many task natural language processing ranging machine translation question answering reduced problem matching two sentence generally two short text propose new approach problem called Deep Match Tree DeepMatchtree general setting approach consists two component 1 mini...
[0.051162995398044586, 0.018305283039808273, -0.01855611428618431, 0.07697848230600357, -0.08832608908414841, -0.0020218202844262123, -0.034851547330617905, 0.02787567302584648, 0.000578019767999649, -0.05271303653717041, 0.005947253666818142, -0.01473609171807766, 0.016358161345124245, 0.0718144103884697, -0.003472416...
524
524
['Grégoire Mesnil', 'Tomas Mikolov', "Marc'Aurelio Ranzato", 'Yoshua Bengio']
1412.5335v7
Sentiment analysis is a common task in natural language processing that aims to detect polarity of a text document (typically a consumer review). In the simplest settings, we discriminate only between positive and negative sentiment, turning the task into a standard binary classification problem. We compare several ma-...
Ensemble of Generative and Discriminative Techniques for Sentiment Analysis of Movie Reviews
2,014
http://arxiv.org/pdf/1412.5335v7
Title Ensemble Generative Discriminative Techniques Sentiment Analysis Movie Reviews Summary Sentiment analysis common task natural language processing aim detect polarity text document typically consumer review simplest setting discriminate positive negative sentiment turning task standard binary classification proble...
[0.040673088282346725, 0.06416871398687363, -0.0010957929771393538, 0.022742290049791336, -0.02854306995868683, 0.02009245753288269, 0.01908349059522152, 0.005958822555840015, -0.006910823751240969, -0.06278390437364578, -0.00593555485829711, -0.03153858333826065, 0.014895033091306686, 0.03587951511144638, -0.027388690...
525
525
['Kartik Audhkhasi', 'Abhinav Sethy', 'Bhuvana Ramabhadran']
1412.7063v5
We propose Diverse Embedding Neural Network (DENN), a novel architecture for language models (LMs). A DENNLM projects the input word history vector onto multiple diverse low-dimensional sub-spaces instead of a single higher-dimensional sub-space as in conventional feed-forward neural network LMs. We encourage these sub...
Diverse Embedding Neural Network Language Models
2,014
http://arxiv.org/pdf/1412.7063v5
Title Diverse Embedding Neural Network Language Models Summary propose Diverse Embedding Neural Network DENN novel architecture language model LMs DENNLM project input word history vector onto multiple diverse lowdimensional subspace instead single higherdimensional subspace conventional feedforward neural network LMs ...
[0.02236238308250904, 0.04809650033712387, -0.011370988562703133, 0.04160626605153084, -0.0006663855165243149, 0.004996190778911114, 0.03278935328125954, -0.01274781208485365, 0.010790622793138027, -0.03399166464805603, -0.0007355916313827038, -0.057956546545028687, 0.020329143851995468, 0.041717637330293655, 0.0400969...
526
526
['Dimitri Palaz', 'Mathew Magimai Doss', 'Ronan Collobert']
1412.7110v6
Automatic speech recognition systems usually rely on spectral-based features, such as MFCC of PLP. These features are extracted based on prior knowledge such as, speech perception or/and speech production. Recently, convolutional neural networks have been shown to be able to estimate phoneme conditional probabilities i...
Learning linearly separable features for speech recognition using convolutional neural networks
2,014
http://arxiv.org/pdf/1412.7110v6
Title Learning linearly separable feature speech recognition using convolutional neural network Summary Automatic speech recognition system usually rely spectralbased feature MFCC PLP feature extracted based prior knowledge speech perception orand speech production Recently convolutional neural network shown able estim...
[0.00563886109739542, 0.029790548607707024, 0.021793151274323463, 0.07984942197799683, 0.009085021913051605, -0.007258158177137375, 0.05035756155848503, 0.01637815125286579, 0.012337390333414078, -0.014996013604104519, -0.040835052728652954, 0.02297447808086872, -0.0020683889742940664, 0.07877500355243683, 0.0013255557...
527
527
['Edward Grefenstette', 'Karl Moritz Hermann', 'Mustafa Suleyman', 'Phil Blunsom']
1506.02516v3
Recently, strong results have been demonstrated by Deep Recurrent Neural Networks on natural language transduction problems. In this paper we explore the representational power of these models using synthetic grammars designed to exhibit phenomena similar to those found in real transduction problems such as machine tra...
Learning to Transduce with Unbounded Memory
2,015
http://arxiv.org/pdf/1506.02516v3
Title Learning Transduce Unbounded Memory Summary Recently strong result demonstrated Deep Recurrent Neural Networks natural language transduction problem paper explore representational power model using synthetic grammar designed exhibit phenomenon similar found real transduction problem machine translation experiment...
[0.020350556820631027, 0.008127058856189251, -0.015597475692629814, 0.05192788690328598, -0.00637725880369544, 0.00024154184211511165, -0.026283787563443184, -0.016088377684354782, -0.04758871719241142, -0.03196248784661293, 0.002205190947279334, -0.05149921402335167, -0.00998506136238575, 0.036861661821603775, 0.02149...
528
528
['ShiLiang Zhang', 'Hui Jiang', 'Si Wei', 'LiRong Dai']
1510.02693v1
We introduce a new structure for memory neural networks, called feedforward sequential memory networks (FSMN), which can learn long-term dependency without using recurrent feedback. The proposed FSMN is a standard feedforward neural networks equipped with learnable sequential memory blocks in the hidden layers. In this...
Feedforward Sequential Memory Neural Networks without Recurrent Feedback
2,015
http://arxiv.org/pdf/1510.02693v1
Title Feedforward Sequential Memory Neural Networks without Recurrent Feedback Summary introduce new structure memory neural network called feedforward sequential memory network FSMN learn longterm dependency without using recurrent feedback proposed FSMN standard feedforward neural network equipped learnable sequentia...
[0.0170398261398077, 0.02921907790005207, 0.0004469997074920684, 0.053692590445280075, -0.015155096538364887, -0.004730031359940767, 0.001771857962012291, -0.0001782434555934742, 0.039970770478248596, -0.026382112875580788, -0.013815529644489288, -0.04723210632801056, 0.022966090589761734, 0.041136160492897034, 0.00424...
529
529
['Yi-Hsiu Liao', 'Hung-yi Lee', 'Lin-shan Lee']
1511.02506v1
In this paper we propose the Structured Deep Neural Network (structured DNN) as a structured and deep learning framework. This approach can learn to find the best structured object (such as a label sequence) given a structured input (such as a vector sequence) by globally considering the mapping relationships between t...
Towards Structured Deep Neural Network for Automatic Speech Recognition
2,015
http://arxiv.org/pdf/1511.02506v1
Title Towards Structured Deep Neural Network Automatic Speech Recognition Summary paper propose Structured Deep Neural Network structured DNN structured deep learning framework approach learn find best structured object label sequence given structured input vector sequence globally considering mapping relationship stru...
[-0.020786505192518234, 0.031145649030804634, 0.023787403479218483, 0.03606473654508591, 0.007402504328638315, -0.011239790357649326, 0.030399184674024582, -0.007328341715037823, 0.006598798092454672, -0.012059276923537254, -0.017939789220690727, -0.0021720367949455976, 0.022716106846928596, 0.07752376049757004, -0.006...
530
530
['Kyuyeon Hwang', 'Wonyong Sung']
1601.06581v2
In real-time speech recognition applications, the latency is an important issue. We have developed a character-level incremental speech recognition (ISR) system that responds quickly even during the speech, where the hypotheses are gradually improved while the speaking proceeds. The algorithm employs a speech-to-charac...
Character-Level Incremental Speech Recognition with Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1601.06581v2
Title CharacterLevel Incremental Speech Recognition Recurrent Neural Networks Summary realtime speech recognition application latency important issue developed characterlevel incremental speech recognition ISR system responds quickly even speech hypothesis gradually improved speaking proceeds algorithm employ speechtoc...
[0.0044223531149327755, 0.05571771040558815, 0.03744976967573166, 0.038522932678461075, -0.008043334819376469, -0.006230231840163469, 0.005893508903682232, -0.0026893592439591885, -0.051229141652584076, -0.003056955523788929, -0.018325883895158768, -0.04175221920013428, 0.06111201271414757, 0.07964888960123062, -0.0042...
531
531
['Daniel Andor', 'Chris Alberti', 'David Weiss', 'Aliaksei Severyn', 'Alessandro Presta', 'Kuzman Ganchev', 'Slav Petrov', 'Michael Collins']
1603.06042v2
We introduce a globally normalized transition-based neural network model that achieves state-of-the-art part-of-speech tagging, dependency parsing and sentence compression results. Our model is a simple feed-forward neural network that operates on a task-specific transition system, yet achieves comparable or better acc...
Globally Normalized Transition-Based Neural Networks
2,016
http://arxiv.org/pdf/1603.06042v2
Title Globally Normalized TransitionBased Neural Networks Summary introduce globally normalized transitionbased neural network model achieves stateoftheart partofspeech tagging dependency parsing sentence compression result model simple feedforward neural network operates taskspecific transition system yet achieves com...
[-0.0018425033194944263, 0.03360210731625557, 0.004456736147403717, 6.53143142699264e-05, -0.022016210481524467, -0.017721550539135933, 0.009323908016085625, -0.012625712901353836, -0.015643484890460968, 0.0013822111068293452, -0.0024533974938094616, 0.024322573095560074, -0.011615287512540817, 0.07487696409225464, -0....
532
532
['Peng Li', 'Heng Huang']
1603.09381v1
We report an implementation of a clinical information extraction tool that leverages deep neural network to annotate event spans and their attributes from raw clinical notes and pathology reports. Our approach uses context words and their part-of-speech tags and shape information as features. Then we hire temporal (1D)...
Clinical Information Extraction via Convolutional Neural Network
2,016
http://arxiv.org/pdf/1603.09381v1
Title Clinical Information Extraction via Convolutional Neural Network Summary report implementation clinical information extraction tool leverage deep neural network annotate event span attribute raw clinical note pathology report approach us context word partofspeech tag shape information feature hire temporal 1D con...
[0.029269743710756302, 0.0399782657623291, 0.013260511681437492, 0.0400710366666317, -0.011499225161969662, 0.03400396928191185, -0.0003108444798272103, 0.03770548105239868, -0.002720797434449196, 0.00375669589266181, -0.0022862101905047894, -0.023347411304712296, -0.009924249723553658, 0.09361836314201355, -0.00325862...
533
533
['David Krueger', 'Tegan Maharaj', 'János Kramár', 'Mohammad Pezeshki', 'Nicolas Ballas', 'Nan Rosemary Ke', 'Anirudh Goyal', 'Yoshua Bengio', 'Aaron Courville', 'Chris Pal']
1606.01305v4
We propose zoneout, a novel method for regularizing RNNs. At each timestep, zoneout stochastically forces some hidden units to maintain their previous values. Like dropout, zoneout uses random noise to train a pseudo-ensemble, improving generalization. But by preserving instead of dropping hidden units, gradient inform...
Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations
2,016
http://arxiv.org/pdf/1606.01305v4
Title Zoneout Regularizing RNNs Randomly Preserving Hidden Activations Summary propose zoneout novel method regularizing RNNs timestep zoneout stochastically force hidden unit maintain previous value Like dropout zoneout us random noise train pseudoensemble improving generalization preserving instead dropping hidden un...
[0.019840899854898453, 0.05786997452378273, 0.011844219639897346, 0.013934220187366009, -0.03447861969470978, -0.015363059937953949, 0.04581072926521301, 0.005135738290846348, -0.029061265289783478, 0.005191651172935963, -0.04723426699638367, -0.0017743544885888696, 0.02767947129905224, 0.02805199846625328, 0.005984567...
534
534
['Isabelle Augenstein', 'Tim Rocktäschel', 'Andreas Vlachos', 'Kalina Bontcheva']
1606.05464v2
Stance detection is the task of classifying the attitude expressed in a text towards a target such as Hillary Clinton to be "positive", negative" or "neutral". Previous work has assumed that either the target is mentioned in the text or that training data for every target is given. This paper considers the more challen...
Stance Detection with Bidirectional Conditional Encoding
2,016
http://arxiv.org/pdf/1606.05464v2
Title Stance Detection Bidirectional Conditional Encoding Summary Stance detection task classifying attitude expressed text towards target Hillary Clinton positive negative neutral Previous work assumed either target mentioned text training data every target given paper considers challenging version task target always ...
[0.024833083152770996, 0.0354040190577507, 0.015504857525229454, 0.05491749569773674, -0.017649494111537933, 0.036230795085430145, -0.014465161599218845, 0.01846902258694172, -0.0023728450760245323, -0.08561266958713531, -0.0020045368000864983, -0.037640057504177094, -0.049539320170879364, 0.03160585090517998, -0.01427...
535
535
['Noura Al Moubayed', 'Toby Breckon', 'Peter Matthews', 'A. Stephen McGough']
1606.05554v1
In This paper we present a novel approach to spam filtering and demonstrate its applicability with respect to SMS messages. Our approach requires minimum features engineering and a small set of la- belled data samples. Features are extracted using topic modelling based on latent Dirichlet allocation, and then a compreh...
SMS Spam Filtering using Probabilistic Topic Modelling and Stacked Denoising Autoencoder
2,016
http://arxiv.org/pdf/1606.05554v1
Title SMS Spam Filtering using Probabilistic Topic Modelling Stacked Denoising Autoencoder Summary paper present novel approach spam filtering demonstrate applicability respect SMS message approach requires minimum feature engineering small set la belled data sample Features extracted using topic modelling based latent...
[0.07865723967552185, -0.014085316099226475, -0.03809920698404312, 0.018198154866695404, -0.04089958220720291, 0.0033303394448012114, 0.027079053223133087, -0.013853617012500763, -0.0011168353958055377, -0.06289788335561752, 0.018436282873153687, 0.0021972579415887594, -0.010859817266464233, 0.07428963482379913, -0.055...
536
536
['Abhyuday Jagannatha', 'Hong Yu']
1606.07953v2
Sequence labeling for extraction of medical events and their attributes from unstructured text in Electronic Health Record (EHR) notes is a key step towards semantic understanding of EHRs. It has important applications in health informatics including pharmacovigilance and drug surveillance. The state of the art supervi...
Bidirectional Recurrent Neural Networks for Medical Event Detection in Electronic Health Records
2,016
http://arxiv.org/pdf/1606.07953v2
Title Bidirectional Recurrent Neural Networks Medical Event Detection Electronic Health Records Summary Sequence labeling extraction medical event attribute unstructured text Electronic Health Record EHR note key step towards semantic understanding EHRs important application health informatics including pharmacovigilan...
[0.028854914009571075, 0.021569009870290756, 0.02300029993057251, 0.0031422856263816357, -0.017468098551034927, 0.02254670299589634, -0.015461267903447151, 0.02820962853729725, 0.019318534061312675, -0.03862966224551201, 0.07021252065896988, -0.047684554010629654, -0.005900430493056774, 0.0902818888425827, -0.012088811...
537
537
['Liang Lu']
1607.01963v5
Highway deep neural network (HDNN) is a type of depth-gated feedforward neural network, which has shown to be easier to train with more hidden layers and also generalise better compared to conventional plain deep neural networks (DNNs). Previously, we investigated a structured HDNN architecture for speech recognition, ...
Sequence Training and Adaptation of Highway Deep Neural Networks
2,016
http://arxiv.org/pdf/1607.01963v5
Title Sequence Training Adaptation Highway Deep Neural Networks Summary Highway deep neural network HDNN type depthgated feedforward neural network shown easier train hidden layer also generalise better compared conventional plain deep neural network DNNs Previously investigated structured HDNN architecture speech reco...
[0.008032679557800293, 0.06332380324602127, -0.0070065963082015514, 0.0451037734746933, 0.0027463044971227646, 0.0019146653357893229, 0.04481321945786476, 0.0075764842331409454, 0.0069970544427633286, -0.04749356210231781, -0.031650710850954056, -0.0027913234662264585, 0.024920636788010597, 0.032488442957401276, 0.0068...
538
538
['Julian Georg Zilly', 'Rupesh Kumar Srivastava', 'Jan Koutník', 'Jürgen Schmidhuber']
1607.03474v5
Many sequential processing tasks require complex nonlinear transition functions from one step to the next. However, recurrent neural networks with 'deep' transition functions remain difficult to train, even when using Long Short-Term Memory (LSTM) networks. We introduce a novel theoretical analysis of recurrent network...
Recurrent Highway Networks
2,016
http://arxiv.org/pdf/1607.03474v5
Title Recurrent Highway Networks Summary Many sequential processing task require complex nonlinear transition function one step next However recurrent neural network deep transition function remain difficult train even using Long ShortTerm Memory LSTM network introduce novel theoretical analysis recurrent network based...
[0.009730732999742031, 0.05000156909227371, -0.009580492973327637, 0.02807185985147953, -0.009045728482306004, 0.009133345447480679, -0.005226182285696268, 0.01703661121428013, -0.022015279158949852, -0.00795978307723999, 0.033901527523994446, -0.0669899731874466, 0.029679475352168083, 0.04942607879638672, -0.007268294...
539
539
['Antonio Valerio Miceli Barone']
1608.02996v1
Current approaches to learning vector representations of text that are compatible between different languages usually require some amount of parallel text, aligned at word, sentence or at least document level. We hypothesize however, that different natural languages share enough semantic structure that it should be pos...
Towards cross-lingual distributed representations without parallel text trained with adversarial autoencoders
2,016
http://arxiv.org/pdf/1608.02996v1
Title Towards crosslingual distributed representation without parallel text trained adversarial autoencoders Summary Current approach learning vector representation text compatible different language usually require amount parallel text aligned word sentence least document level hypothesize however different natural la...
[-0.002654540818184614, 0.029592353850603104, -0.020501552149653435, 0.08953430503606796, -0.05177861452102661, 0.015171200037002563, 0.030116500332951546, -0.03203490003943443, 0.01959400065243244, -0.04761359840631485, -0.051239270716905594, -0.00202724477276206, 0.045413024723529816, -0.007184742484241724, 0.0145744...
540
540
['Zhiyuan Tang', 'Ying Shi', 'Dong Wang', 'Yang Feng', 'Shiyue Zhang']
1609.08789v3
Recurrent neural networks (RNNs) have shown clear superiority in sequence modeling, particularly the ones with gated units, such as long short-term memory (LSTM) and gated recurrent unit (GRU). However, the dynamic properties behind the remarkable performance remain unclear in many applications, e.g., automatic speech ...
Memory Visualization for Gated Recurrent Neural Networks in Speech Recognition
2,016
http://arxiv.org/pdf/1609.08789v3
Title Memory Visualization Gated Recurrent Neural Networks Speech Recognition Summary Recurrent neural network RNNs shown clear superiority sequence modeling particularly one gated unit long shortterm memory LSTM gated recurrent unit GRU However dynamic property behind remarkable performance remain unclear many applica...
[-0.006197511218488216, 0.0016232029302045703, -0.010341375134885311, 0.04599903151392937, -0.024055778980255127, -0.005096917040646076, 0.060437288135290146, -0.02678726799786091, -0.015349045395851135, 0.0017848003190010786, -0.032852910459041595, -0.04186113923788071, 0.05244435369968414, 0.05691493675112724, 0.0358...
541
541
['Hagen Soltau', 'Hank Liao', 'Hasim Sak']
1610.09975v1
We present results that show it is possible to build a competitive, greatly simplified, large vocabulary continuous speech recognition system with whole words as acoustic units. We model the output vocabulary of about 100,000 words directly using deep bi-directional LSTM RNNs with CTC loss. The model is trained on 125,...
Neural Speech Recognizer: Acoustic-to-Word LSTM Model for Large Vocabulary Speech Recognition
2,016
http://arxiv.org/pdf/1610.09975v1
Title Neural Speech Recognizer AcoustictoWord LSTM Model Large Vocabulary Speech Recognition Summary present result show possible build competitive greatly simplified large vocabulary continuous speech recognition system whole word acoustic unit model output vocabulary 100000 word directly using deep bidirectional LSTM...
[-0.004185554105788469, 0.043656352907419205, 0.04244375228881836, 0.09234080463647842, -0.003420830937102437, 0.00681381905451417, 0.03762492164969444, 0.013959381729364395, -0.03840254992246628, -0.021074889227747917, -0.0640619620680809, -0.03905227780342102, 0.03335243836045265, 0.04141192510724068, 0.0106396749615...
542
542
['Prajit Ramachandran', 'Peter J. Liu', 'Quoc V. Le']
1611.02683v2
This work presents a general unsupervised learning method to improve the accuracy of sequence to sequence (seq2seq) models. In our method, the weights of the encoder and decoder of a seq2seq model are initialized with the pretrained weights of two language models and then fine-tuned with labeled data. We apply this met...
Unsupervised Pretraining for Sequence to Sequence Learning
2,016
http://arxiv.org/pdf/1611.02683v2
Title Unsupervised Pretraining Sequence Sequence Learning Summary work present general unsupervised learning method improve accuracy sequence sequence seq2seq model method weight encoder decoder seq2seq model initialized pretrained weight two language model finetuned labeled data apply method challenging benchmark mach...
[0.002003306522965431, 0.06992283463478088, -0.011045689694583416, 0.01799277774989605, -0.0049300226382911205, 0.02622196078300476, 0.021231332793831825, -0.005262626335024834, -0.029905002564191818, -0.045856647193431854, -0.029438117519021034, -0.013366500847041607, 0.030526725575327873, 0.03710099309682846, 0.01927...
543
543
['Yoon Kim', 'Carl Denton', 'Luong Hoang', 'Alexander M. Rush']
1702.00887v3
Attention networks have proven to be an effective approach for embedding categorical inference within a deep neural network. However, for many tasks we may want to model richer structural dependencies without abandoning end-to-end training. In this work, we experiment with incorporating richer structural distributions,...
Structured Attention Networks
2,017
http://arxiv.org/pdf/1702.00887v3
Title Structured Attention Networks Summary Attention network proven effective approach embedding categorical inference within deep neural network However many task may want model richer structural dependency without abandoning endtoend training work experiment incorporating richer structural distribution encoded using...
[0.026088450103998184, 0.0376410074532032, -0.01531126070767641, 0.04309573397040367, -0.001533530536107719, -0.003978844732046127, -0.0019318856066092849, -0.026550261303782463, -0.006426883861422539, -0.05096250772476196, -0.009327799081802368, 0.032912496477365494, -0.0024358744267374277, 0.05665376037359238, 0.0270...
544
544
['Hongyu Guo', 'Colin Cherry', 'Jiang Su']
1704.05907v1
We propose a multi-view network for text classification. Our method automatically creates various views of its input text, each taking the form of soft attention weights that distribute the classifier's focus among a set of base features. For a bag-of-words representation, each view focuses on a different subset of the...
End-to-End Multi-View Networks for Text Classification
2,017
http://arxiv.org/pdf/1704.05907v1
Title EndtoEnd MultiView Networks Text Classification Summary propose multiview network text classification method automatically creates various view input text taking form soft attention weight distribute classifier focus among set base feature bagofwords representation view focus different subset text word Aggregatin...
[0.025872070342302322, -0.006213345564901829, 0.024412499740719795, 0.04621824249625206, -0.030140504240989685, 0.004059010650962591, 0.03245772048830986, 0.012296872213482857, 0.04079432040452957, -0.06887512654066086, -0.033552028238773346, -0.011995146051049232, 0.011675628833472729, 0.04331379756331444, 0.003311428...
545
545
['Kartik Goyal', 'Chris Dyer', 'Taylor Berg-Kirkpatrick']
1704.06970v1
We demonstrate that a continuous relaxation of the argmax operation can be used to create a differentiable approximation to greedy decoding for sequence-to-sequence (seq2seq) models. By incorporating this approximation into the scheduled sampling training procedure (Bengio et al., 2015)--a well-known technique for corr...
Differentiable Scheduled Sampling for Credit Assignment
2,017
http://arxiv.org/pdf/1704.06970v1
Title Differentiable Scheduled Sampling Credit Assignment Summary demonstrate continuous relaxation argmax operation used create differentiable approximation greedy decoding sequencetosequence seq2seq model incorporating approximation scheduled sampling training procedure Bengio et al 2015a wellknown technique correcti...
[0.04359620064496994, 0.042325232177972794, 0.006403833162039518, 0.015020525082945824, 0.009850161150097847, 0.05608697980642319, -0.016278481110930443, 0.005217379424721003, -0.014566252939403057, -0.017947543412446976, -0.022926131263375282, -0.0151705052703619, 0.018968705087900162, -0.028447363525629044, 0.0015770...
546
546
['Zhiyuan Tang', 'Dong Wang', 'Yixiang Chen', 'Ying Shi', 'Lantian Li']
1705.03152v2
Pure acoustic neural models, particularly the LSTM-RNN model, have shown great potential in language identification (LID). However, the phonetic information has been largely overlooked by most of existing neural LID models, although this information has been used in the conventional phonetic LID systems with a great su...
Phone-aware Neural Language Identification
2,017
http://arxiv.org/pdf/1705.03152v2
Title Phoneaware Neural Language Identification Summary Pure acoustic neural model particularly LSTMRNN model shown great potential language identification LID However phonetic information largely overlooked existing neural LID model although information used conventional phonetic LID system great success present phone...
[0.01298956386744976, 0.08186398446559906, 0.015031697228550911, 0.04583071172237396, -0.013382386416196823, 0.01289944164454937, 0.055056098848581314, -0.01047565694898367, -0.016969528049230576, -0.0007712991209700704, -0.06293878704309464, -0.04870281741023064, 0.05759785696864128, 0.0004911783616989851, 0.043796192...
547
547
['Marek Rei']
1707.05233v1
Automated methods for essay scoring have made great progress in recent years, achieving accuracies very close to human annotators. However, a known weakness of such automated scorers is not taking into account the semantic relevance of the submitted text. While there is existing work on detecting answer relevance given...
Detecting Off-topic Responses to Visual Prompts
2,017
http://arxiv.org/pdf/1707.05233v1
Title Detecting Offtopic Responses Visual Prompts Summary Automated method essay scoring made great progress recent year achieving accuracy close human annotator However known weakness automated scorer taking account semantic relevance submitted text existing work detecting answer relevance given textual prompt little ...
[0.10225941240787506, -0.0015375548973679543, -0.01723756641149521, 0.03070058673620224, -0.0012213430600240827, 0.011814624071121216, 0.002692685928195715, 0.06981052458286285, 0.017407312989234924, -0.04312202334403992, 0.012537005357444286, 0.007268738467246294, -0.010276996530592442, 0.06191529333591461, -0.0303712...
548
548
['Fréderic Godin', 'Jonas Degrave', 'Joni Dambre', 'Wesley De Neve']
1707.08214v2
In this paper, we introduce a novel type of Rectified Linear Unit (ReLU), called a Dual Rectified Linear Unit (DReLU). A DReLU, which comes with an unbounded positive and negative image, can be used as a drop-in replacement for a tanh activation function in the recurrent step of Quasi-Recurrent Neural Networks (QRNNs) ...
Dual Rectified Linear Units (DReLUs): A Replacement for Tanh Activation Functions in Quasi-Recurrent Neural Networks
2,017
http://arxiv.org/pdf/1707.08214v2
Title Dual Rectified Linear Units DReLUs Replacement Tanh Activation Functions QuasiRecurrent Neural Networks Summary paper introduce novel type Rectified Linear Unit ReLU called Dual Rectified Linear Unit DReLU DReLU come unbounded positive negative image used dropin replacement tanh activation function recurrent step...
[0.0021918551065027714, 0.008251101709902287, 0.01522910874336958, 0.04784897714853287, -0.017525488510727882, 0.005926362704485655, 0.02409275807440281, -0.007333395536988974, -0.02322854846715927, -0.036688949912786484, -0.053567834198474884, -0.06477624177932739, 0.007544497959315777, 0.07392292469739914, -0.0106450...
549
549
['Mostafa Dehghani', 'Arash Mehrjou', 'Stephan Gouws', 'Jaap Kamps', 'Bernhard Schölkopf']
1711.02799v1
Training deep neural networks requires many training samples, but in practice training labels are expensive to obtain and may be of varying quality, as some may be from trusted expert labelers while others might be from heuristics or other sources of weak supervision such as crowd-sourcing. This creates a fundamental q...
Fidelity-Weighted Learning
2,017
http://arxiv.org/pdf/1711.02799v1
Title FidelityWeighted Learning Summary Training deep neural network requires many training sample practice training label expensive obtain may varying quality may trusted expert labelers others might heuristic source weak supervision crowdsourcing creates fundamental quality versusquantity tradeoff learning process le...
[0.05056807026267052, 0.05870668217539787, -0.02262929268181324, 0.04009771719574928, 0.021441899240016937, -0.03986431658267975, 0.03014550544321537, -0.03484506532549858, -0.014632077887654305, -0.01139342412352562, -0.04413081333041191, -0.0247972309589386, 0.007972529157996178, 0.057504575699567795, 0.0095823891460...
550
550
['Dong Yu', 'Michael L. Seltzer', 'Jinyu Li', 'Jui-Ting Huang', 'Frank Seide']
1301.3605v3
Recent studies have shown that deep neural networks (DNNs) perform significantly better than shallow networks and Gaussian mixture models (GMMs) on large vocabulary speech recognition tasks. In this paper, we argue that the improved accuracy achieved by the DNNs is the result of their ability to extract discriminative ...
Feature Learning in Deep Neural Networks - Studies on Speech Recognition Tasks
2,013
http://arxiv.org/pdf/1301.3605v3
Title Feature Learning Deep Neural Networks Studies Speech Recognition Tasks Summary Recent study shown deep neural network DNNs perform significantly better shallow network Gaussian mixture model GMMs large vocabulary speech recognition task paper argue improved accuracy achieved DNNs result ability extract discrimina...
[-0.017452381551265717, -0.002058636164292693, 0.0036066260654479265, 0.03968363255262375, -0.007213184610009193, -0.017433112487196922, 0.05981472134590149, -0.017009399831295013, -0.04588581249117851, 0.01918136700987816, -0.041604217141866684, 0.01103188656270504, 0.04672212898731232, 0.03411168232560158, 0.01186550...
551
551
['Dimitri Palaz', 'Ronan Collobert', 'Mathew Magimai. -Doss']
1304.1018v2
In hybrid hidden Markov model/artificial neural networks (HMM/ANN) automatic speech recognition (ASR) system, the phoneme class conditional probabilities are estimated by first extracting acoustic features from the speech signal based on prior knowledge such as, speech perception or/and speech production knowledge, and...
Estimating Phoneme Class Conditional Probabilities from Raw Speech Signal using Convolutional Neural Networks
2,013
http://arxiv.org/pdf/1304.1018v2
Title Estimating Phoneme Class Conditional Probabilities Raw Speech Signal using Convolutional Neural Networks Summary hybrid hidden Markov modelartificial neural network HMMANN automatic speech recognition ASR system phoneme class conditional probability estimated first extracting acoustic feature speech signal based ...
[0.013593069277703762, 0.04590294510126114, 0.017292046919465065, 0.0597132183611393, 0.004294393118470907, -0.019776178523898125, 0.04187477007508278, 0.006013117264956236, -0.034289129078388214, -0.03814668580889702, -0.0364546962082386, -0.013362758792936802, 0.01716701127588749, 0.03846025466918945, 0.0041391011327...
552
552
['Samuel R. Bowman', 'Christopher Potts', 'Christopher D. Manning']
1406.1827v4
Tree-structured recursive neural networks (TreeRNNs) for sentence meaning have been successful for many applications, but it remains an open question whether the fixed-length representations that they learn can support tasks as demanding as logical deduction. We pursue this question by evaluating whether two such model...
Recursive Neural Networks Can Learn Logical Semantics
2,014
http://arxiv.org/pdf/1406.1827v4
Title Recursive Neural Networks Learn Logical Semantics Summary Treestructured recursive neural network TreeRNNs sentence meaning successful many application remains open question whether fixedlength representation learn support task demanding logical deduction pursue question evaluating whether two modelsplain TreeRNN...
[0.04599861800670624, 0.04760322719812393, -0.021887384355068207, 0.05416524410247803, -0.05406181514263153, 0.007405961398035288, -0.010208075866103172, -0.01070459932088852, 0.0029791484121233225, -0.04488600790500641, 0.039833322167396545, 0.03173527494072914, 0.006863091140985489, 0.051395632326602936, 0.0032243358...
553
553
['Chenxi Zhu', 'Xipeng Qiu', 'Xinchi Chen', 'Xuanjing Huang']
1505.05667v1
In this work, we address the problem to model all the nodes (words or phrases) in a dependency tree with the dense representations. We propose a recursive convolutional neural network (RCNN) architecture to capture syntactic and compositional-semantic representations of phrases and words in a dependency tree. Different...
A Re-ranking Model for Dependency Parser with Recursive Convolutional Neural Network
2,015
http://arxiv.org/pdf/1505.05667v1
Title Reranking Model Dependency Parser Recursive Convolutional Neural Network Summary work address problem model node word phrase dependency tree dense representation propose recursive convolutional neural network RCNN architecture capture syntactic compositionalsemantic representation phrase word dependency tree Diff...
[0.058216795325279236, 0.036891523748636246, 0.007242720574140549, 0.09531322866678238, -0.04551072418689728, -2.6305309802410193e-05, -0.04743984714150429, -0.008480262942612171, -0.007531933486461639, -0.03914311155676842, 0.0289884265512228, 0.016147201880812645, 0.030944496393203735, 0.022281188517808914, -0.046962...
554
554
['Lantian Li', 'Dong Wang', 'Zhiyong Zhang', 'Thomas Fang Zheng']
1505.06427v1
Recent research shows that deep neural networks (DNNs) can be used to extract deep speaker vectors (d-vectors) that preserve speaker characteristics and can be used in speaker verification. This new method has been tested on text-dependent speaker verification tasks, and improvement was reported when combined with the ...
Deep Speaker Vectors for Semi Text-independent Speaker Verification
2,015
http://arxiv.org/pdf/1505.06427v1
Title Deep Speaker Vectors Semi Textindependent Speaker Verification Summary Recent research show deep neural network DNNs used extract deep speaker vector dvectors preserve speaker characteristic used speaker verification new method tested textdependent speaker verification task improvement reported combined conventio...
[-0.004094146192073822, 0.025912322103977203, -0.00012574568972922862, 0.06336522102355957, -0.013480176217854023, 0.013753141276538372, 0.05886764079332352, -0.011309025809168816, -0.015574604272842407, -0.005354780238121748, -0.0016111833974719048, -0.03165379539132118, 0.02726726606488228, -0.008409736678004265, 0.0...
555
555
['Tom Sercu', 'Vaibhava Goel']
1604.01792v2
Very deep CNNs with small 3x3 kernels have recently been shown to achieve very strong performance as acoustic models in hybrid NN-HMM speech recognition systems. In this paper we investigate how to efficiently scale these models to larger datasets. Specifically, we address the design choice of pooling and padding along...
Advances in Very Deep Convolutional Neural Networks for LVCSR
2,016
http://arxiv.org/pdf/1604.01792v2
Title Advances Deep Convolutional Neural Networks LVCSR Summary deep CNNs small 3x3 kernel recently shown achieve strong performance acoustic model hybrid NNHMM speech recognition system paper investigate efficiently scale model larger datasets Specifically address design choice pooling padding along time dimension ren...
[-0.0027890349738299847, 0.029791463166475296, 0.027223581448197365, 0.0857696682214737, 0.010207554325461388, -0.03195658326148987, 0.05236981064081192, -0.02051599696278572, -0.027545113116502762, 0.01005551777780056, -0.06834647059440613, -0.022901779040694237, 0.012852547690272331, 0.035293564200401306, 0.023511055...
556
556
['Zhiyun Lu', 'Vikas Sindhwani', 'Tara N. Sainath']
1604.02594v1
Recurrent neural networks (RNNs), including long short-term memory (LSTM) RNNs, have produced state-of-the-art results on a variety of speech recognition tasks. However, these models are often too large in size for deployment on mobile devices with memory and latency constraints. In this work, we study mechanisms for l...
Learning Compact Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1604.02594v1
Title Learning Compact Recurrent Neural Networks Summary Recurrent neural network RNNs including long shortterm memory LSTM RNNs produced stateoftheart result variety speech recognition task However model often large size deployment mobile device memory latency constraint work study mechanism learning compact RNNs LSTM...
[-0.0004871900600846857, -0.004935301840305328, 0.0123648876324296, 0.05660497397184372, 0.00821257196366787, -0.031614307314157486, 0.04056230187416077, -0.0007642898708581924, -0.02976626344025135, -0.010106392204761505, -0.04787039756774902, -0.03563445433974266, 0.03354519233107567, -0.0073138996958732605, -0.01384...
557
557
['Adhiguna Kuncoro', 'Yuichiro Sawai', 'Kevin Duh', 'Yuji Matsumoto']
1604.06529v2
We propose a transition-based dependency parser using Recurrent Neural Networks with Long Short-Term Memory (LSTM) units. This extends the feedforward neural network parser of Chen and Manning (2014) and enables modelling of entire sequences of shift/reduce transition decisions. On the Google Web Treebank, our LSTM par...
Dependency Parsing with LSTMs: An Empirical Evaluation
2,016
http://arxiv.org/pdf/1604.06529v2
Title Dependency Parsing LSTMs Empirical Evaluation Summary propose transitionbased dependency parser using Recurrent Neural Networks Long ShortTerm Memory LSTM unit extends feedforward neural network parser Chen Manning 2014 enables modelling entire sequence shiftreduce transition decision Google Web Treebank LSTM par...
[0.029366113245487213, 0.05464070290327072, -0.003604850498959422, 0.047632377594709396, -0.03991984203457832, -0.0028011994436383247, -0.025996096432209015, -0.013133931905031204, -0.025421075522899628, -0.042501360177993774, 0.03368372842669487, -0.010501038283109665, 0.012196731753647327, 0.03819626569747925, -0.043...
558
558
['Hamid Palangi', 'Li Deng', 'Yelong Shen', 'Jianfeng Gao', 'Xiaodong He', 'Jianshu Chen', 'Xinying Song', 'Rabab Ward']
1502.06922v3
This paper develops a model that addresses sentence embedding, a hot topic in current natural language processing research, using recurrent neural networks with Long Short-Term Memory (LSTM) cells. Due to its ability to capture long term memory, the LSTM-RNN accumulates increasingly richer information as it goes throug...
Deep Sentence Embedding Using Long Short-Term Memory Networks: Analysis and Application to Information Retrieval
2,015
http://arxiv.org/pdf/1502.06922v3
Title Deep Sentence Embedding Using Long ShortTerm Memory Networks Analysis Application Information Retrieval Summary paper develops model address sentence embedding hot topic current natural language processing research using recurrent neural network Long ShortTerm Memory LSTM cell Due ability capture long term memory...
[0.03156965970993042, -0.04089318588376045, -0.008375040255486965, 0.05650141090154648, -0.025587432086467743, 0.006287709344178438, 0.002414876129478216, -0.021042458713054657, 0.010764251463115215, -0.06723520904779434, -0.004185568541288376, 0.003493493190035224, -0.010016089305281639, 0.012699143029749393, 0.018240...
559
559
['Fandong Meng', 'Zhengdong Lu', 'Mingxuan Wang', 'Hang Li', 'Wenbin Jiang', 'Qun Liu']
1503.01838v5
The recently proposed neural network joint model (NNJM) (Devlin et al., 2014) augments the n-gram target language model with a heuristically chosen source context window, achieving state-of-the-art performance in SMT. In this paper, we give a more systematic treatment by summarizing the relevant source information thro...
Encoding Source Language with Convolutional Neural Network for Machine Translation
2,015
http://arxiv.org/pdf/1503.01838v5
Title Encoding Source Language Convolutional Neural Network Machine Translation Summary recently proposed neural network joint model NNJM Devlin et al 2014 augments ngram target language model heuristically chosen source context window achieving stateoftheart performance SMT paper give systematic treatment summarizing ...
[0.05548024922609329, 0.047425027936697006, 0.014198657125234604, 0.07506497204303741, -0.03541373834013939, 0.012434015981853008, 0.039030659943819046, 0.004686171654611826, -0.07581836730241776, -0.029282188042998314, -0.05318759009242058, -0.02025393396615982, 0.005503562744706869, 0.04283221811056137, 0.02397718653...
560
560
['Zhen Huang', 'Sabato Marco Siniscalchi', 'I-Fan Chen', 'Jiadong Wu', 'Chin-Hui Lee']
1503.02108v2
We present a Bayesian approach to adapting parameters of a well-trained context-dependent, deep-neural-network, hidden Markov model (CD-DNN-HMM) to improve automatic speech recognition performance. Given an abundance of DNN parameters but with only a limited amount of data, the effectiveness of the adapted DNN model ca...
Maximum a Posteriori Adaptation of Network Parameters in Deep Models
2,015
http://arxiv.org/pdf/1503.02108v2
Title Maximum Posteriori Adaptation Network Parameters Deep Models Summary present Bayesian approach adapting parameter welltrained contextdependent deepneuralnetwork hidden Markov model CDDNNHMM improve automatic speech recognition performance Given abundance DNN parameter limited amount data effectiveness adapted DNN...
[0.016376910731196404, 0.034700606018304825, -0.0065205697901546955, 0.003025778802111745, 0.021287813782691956, 0.020427532494068146, 0.04004742205142975, -0.012555258348584175, -0.020050641149282455, -0.05148341506719589, -0.06654325127601624, 0.03062780201435089, 0.030081646516919136, 0.010403763502836227, 0.0018954...
561
561
['Zhaopeng Tu', 'Baotian Hu', 'Zhengdong Lu', 'Hang Li']
1503.02357v2
We propose a novel method for translation selection in statistical machine translation, in which a convolutional neural network is employed to judge the similarity between a phrase pair in two languages. The specifically designed convolutional architecture encodes not only the semantic similarity of the translation pai...
Context-Dependent Translation Selection Using Convolutional Neural Network
2,015
http://arxiv.org/pdf/1503.02357v2
Title ContextDependent Translation Selection Using Convolutional Neural Network Summary propose novel method translation selection statistical machine translation convolutional neural network employed judge similarity phrase pair two language specifically designed convolutional architecture encodes semantic similarity ...
[0.057121142745018005, 0.03913121297955513, -0.013562625274062157, 0.09125379472970963, -0.055765941739082336, 0.009166888892650604, 0.0394693985581398, 0.009660053066909313, -0.05678289011120796, -0.04626738280057907, -0.053807344287633896, 0.005128434393554926, 0.017527494579553604, -0.03251349925994873, 0.0097431940...
562
562
['Baotian Hu', 'Zhengdong Lu', 'Hang Li', 'Qingcai Chen']
1503.03244v1
Semantic matching is of central importance to many natural language tasks \cite{bordes2014semantic,RetrievalQA}. A successful matching algorithm needs to adequately model the internal structures of language objects and the interaction between them. As a step toward this goal, we propose convolutional neural network mod...
Convolutional Neural Network Architectures for Matching Natural Language Sentences
2,015
http://arxiv.org/pdf/1503.03244v1
Title Convolutional Neural Network Architectures Matching Natural Language Sentences Summary Semantic matching central importance many natural language task citebordes2014semanticRetrievalQA successful matching algorithm need adequately model internal structure language object interaction step toward goal propose convo...
[0.07127247750759125, 0.04071514308452606, 7.114599429769441e-05, 0.10387472808361053, -0.0757235735654831, -0.0007980315713211894, -0.02864399179816246, 0.0014175277901813388, -0.009861437603831291, -0.039536718279123306, 0.001107489806599915, 0.0004916680045425892, 0.007699381094425917, 0.038788337260484695, 0.008469...
563
563
['Xiaodan Zhu', 'Parinaz Sobhani', 'Hongyu Guo']
1503.04881v1
The chain-structured long short-term memory (LSTM) has showed to be effective in a wide range of problems such as speech recognition and machine translation. In this paper, we propose to extend it to tree structures, in which a memory cell can reflect the history memories of multiple child cells or multiple descendant ...
Long Short-Term Memory Over Tree Structures
2,015
http://arxiv.org/pdf/1503.04881v1
Title Long ShortTerm Memory Tree Structures Summary chainstructured long shortterm memory LSTM showed effective wide range problem speech recognition machine translation paper propose extend tree structure memory cell reflect history memory multiple child cell multiple descendant cell recursive process call model SLSTM...
[0.021890904754400253, -0.0186021588742733, 0.0105922631919384, 0.08318793773651123, -0.003978133667260408, 0.0044142757542431355, -6.259512883843854e-05, -0.010849968530237675, -0.04144003242254257, -0.06341680139303207, -0.040655747056007385, -0.06472526490688324, 0.057167090475559235, 0.030477045103907585, -0.018740...
564
564
['Krupakar Hans', 'R S Milton']
1612.02482v2
The advent of the attention mechanism in neural machine translation models has improved the performance of machine translation systems by enabling selective lookup into the source sentence. In this paper, the efficiencies of translation using bidirectional encoder attention decoder models were studied with respect to t...
Improving the Performance of Neural Machine Translation Involving Morphologically Rich Languages
2,016
http://arxiv.org/pdf/1612.02482v2
Title Improving Performance Neural Machine Translation Involving Morphologically Rich Languages Summary advent attention mechanism neural machine translation model improved performance machine translation system enabling selective lookup source sentence paper efficiency translation using bidirectional encoder attention...
[0.06208235025405884, 0.00661317678168416, -0.010848728008568287, 0.06465115398168564, -0.039646659046411514, -0.007100379094481468, -0.01967349275946617, 0.021981865167617798, -0.09096962958574295, -0.04573608934879303, -0.03098827600479126, -0.046243924647569656, 0.05132283642888069, 0.01270466111600399, 0.0617354027...
565
565
['Thomas Laurent', 'James von Brecht']
1612.06212v1
We introduce an exceptionally simple gated recurrent neural network (RNN) that achieves performance comparable to well-known gated architectures, such as LSTMs and GRUs, on the word-level language modeling task. We prove that our model has simple, predicable and non-chaotic dynamics. This stands in stark contrast to mo...
A recurrent neural network without chaos
2,016
http://arxiv.org/pdf/1612.06212v1
Title recurrent neural network without chaos Summary introduce exceptionally simple gated recurrent neural network RNN achieves performance comparable wellknown gated architecture LSTMs GRUs wordlevel language modeling task prove model simple predicable nonchaotic dynamic stand stark contrast standard gated architectur...
[-0.014985080808401108, 0.029927542433142662, -0.028112288564443588, 0.030175624415278435, -0.020678887143731117, -0.00834645051509142, -0.011060236021876335, -0.013258378021419048, -0.03712480887770653, -0.0036131013184785843, 0.022630540654063225, -0.04988980293273926, -0.00441202474758029, 0.07946450263261795, 0.018...
566
566
['Dimitri Palaz', 'Ronan Collobert', 'Mathew Magimai. -Doss']
1312.2137v1
Most phoneme recognition state-of-the-art systems rely on a classical neural network classifiers, fed with highly tuned features, such as MFCC or PLP features. Recent advances in ``deep learning'' approaches questioned such systems, but while some attempts were made with simpler features such as spectrograms, state-of-...
End-to-end Phoneme Sequence Recognition using Convolutional Neural Networks
2,013
http://arxiv.org/pdf/1312.2137v1
Title Endtoend Phoneme Sequence Recognition using Convolutional Neural Networks Summary phoneme recognition stateoftheart system rely classical neural network classifier fed highly tuned feature MFCC PLP feature Recent advance deep learning approach questioned system attempt made simpler feature spectrogram stateofthea...
[0.026019861921668053, 0.06793640553951263, 0.024221213534474373, 0.06060829758644104, 0.027038084343075752, 0.00023603998124599457, 0.04196571558713913, 0.021477218717336655, -0.02096324786543846, 0.010637312196195126, -0.028192345052957535, -0.03161724656820297, 0.03874414414167404, 0.05420498922467232, 0.01417389232...
567
567
['Prasanna Kumar Muthukumar', 'Alan W. Black']
1409.8558v1
Nearly all Statistical Parametric Speech Synthesizers today use Mel Cepstral coefficients as the vocal tract parameterization of the speech signal. Mel Cepstral coefficients were never intended to work in a parametric speech synthesis framework, but as yet, there has been little success in creating a better parameteriz...
A Deep Learning Approach to Data-driven Parameterizations for Statistical Parametric Speech Synthesis
2,014
http://arxiv.org/pdf/1409.8558v1
Title Deep Learning Approach Datadriven Parameterizations Statistical Parametric Speech Synthesis Summary Nearly Statistical Parametric Speech Synthesizers today use Mel Cepstral coefficient vocal tract parameterization speech signal Mel Cepstral coefficient never intended work parametric speech synthesis framework yet...
[-0.026742789894342422, 0.026720818132162094, 0.012580730952322483, -0.01615588739514351, 0.0337187722325325, -0.043105125427246094, 0.045954830944538116, -0.02139735035598278, -0.08642750978469849, 0.004690163303166628, -0.029437372460961342, -0.00020457124628592283, 0.02073962427675724, 0.057533204555511475, 0.040730...
568
568
['Minh-Thang Luong', 'Ilya Sutskever', 'Quoc V. Le', 'Oriol Vinyals', 'Wojciech Zaremba']
1410.8206v4
Neural Machine Translation (NMT) is a new approach to machine translation that has shown promising results that are comparable to traditional approaches. A significant weakness in conventional NMT systems is their inability to correctly translate very rare words: end-to-end NMTs tend to have relatively small vocabulari...
Addressing the Rare Word Problem in Neural Machine Translation
2,014
http://arxiv.org/pdf/1410.8206v4
Title Addressing Rare Word Problem Neural Machine Translation Summary Neural Machine Translation NMT new approach machine translation shown promising result comparable traditional approach significant weakness conventional NMT system inability correctly translate rare word endtoend NMTs tend relatively small vocabulary...
[0.022855162620544434, 0.029289616271853447, 0.02399652637541294, 0.04400232434272766, -0.05045294761657715, -0.023506755009293556, -0.010444186627864838, 0.032771993428468704, -0.04892449453473091, -0.015761377289891243, 0.039404843002557755, -0.015806181356310844, 0.03838327154517174, 0.025523344054818153, 0.03211229...
569
569
['Jianpeng Cheng', 'Dimitri Kartsaklis', 'Edward Grefenstette']
1411.4116v1
This paper aims to explore the effect of prior disambiguation on neural network- based compositional models, with the hope that better semantic representations for text compounds can be produced. We disambiguate the input word vectors before they are fed into a compositional deep net. A series of evaluations shows the ...
Investigating the Role of Prior Disambiguation in Deep-learning Compositional Models of Meaning
2,014
http://arxiv.org/pdf/1411.4116v1
Title Investigating Role Prior Disambiguation Deeplearning Compositional Models Meaning Summary paper aim explore effect prior disambiguation neural network based compositional model hope better semantic representation text compound produced disambiguate input word vector fed compositional deep net series evaluation sh...
[0.05237410217523575, -0.0034892878029495478, -0.017736071720719337, 0.020880402997136116, -0.019608888775110245, 0.031066931784152985, 0.004911269526928663, -0.014840880408883095, -0.01773572526872158, -0.04323531314730644, 0.010033187456429005, -0.01480183657258749, 0.031577616930007935, -0.017169419676065445, 0.0334...
570
570
['Awni Hannun', 'Carl Case', 'Jared Casper', 'Bryan Catanzaro', 'Greg Diamos', 'Erich Elsen', 'Ryan Prenger', 'Sanjeev Satheesh', 'Shubho Sengupta', 'Adam Coates', 'Andrew Y. Ng']
1412.5567v2
We present a state-of-the-art speech recognition system developed using end-to-end deep learning. Our architecture is significantly simpler than traditional speech systems, which rely on laboriously engineered processing pipelines; these traditional systems also tend to perform poorly when used in noisy environments. I...
Deep Speech: Scaling up end-to-end speech recognition
2,014
http://arxiv.org/pdf/1412.5567v2
Title Deep Speech Scaling endtoend speech recognition Summary present stateoftheart speech recognition system developed using endtoend deep learning architecture significantly simpler traditional speech system rely laboriously engineered processing pipeline traditional system also tend perform poorly used noisy environ...
[-0.027821414172649384, -0.007224636618047953, 0.012885487638413906, 0.011362634599208832, 0.05090915039181709, -0.010874909348785877, 0.040681254118680954, 0.012056133709847927, -0.026008332148194313, 0.02655702270567417, -0.055612713098526, -0.03361864015460014, 0.03383137658238411, 0.07654588669538498, 0.01008385419...
571
571
['Aram Ter-Sarkisov', 'Holger Schwenk', 'Loic Barrault', 'Fethi Bougares']
1412.6650v4
It is today acknowledged that neural network language models outperform backoff language models in applications like speech recognition or statistical machine translation. However, training these models on large amounts of data can take several days. We present efficient techniques to adapt a neural network language mo...
Incremental Adaptation Strategies for Neural Network Language Models
2,014
http://arxiv.org/pdf/1412.6650v4
Title Incremental Adaptation Strategies Neural Network Language Models Summary today acknowledged neural network language model outperform backoff language model application like speech recognition statistical machine translation However training model large amount data take several day present efficient technique adap...
[0.04433417692780495, 0.026261765509843826, 0.00290094455704093, 0.009626961313188076, 0.0018505413318052888, 0.007606645580381155, 0.011134685017168522, 0.008617978543043137, -0.029768086969852448, -0.0508238859474659, -0.028443163260817528, -0.0031198954675346613, 0.036626119166612625, 0.0071296426467597485, 0.022294...
572
572
['Joël Legrand', 'Ronan Collobert']
1412.7028v4
This paper introduces a greedy parser based on neural networks, which leverages a new compositional sub-tree representation. The greedy parser and the compositional procedure are jointly trained, and tightly depends on each-other. The composition procedure outputs a vector representation which summarizes syntactically ...
Joint RNN-Based Greedy Parsing and Word Composition
2,014
http://arxiv.org/pdf/1412.7028v4
Title Joint RNNBased Greedy Parsing Word Composition Summary paper introduces greedy parser based neural network leverage new compositional subtree representation greedy parser compositional procedure jointly trained tightly depends eachother composition procedure output vector representation summarizes syntactically p...
[0.038988612592220306, 0.04623155668377876, -0.0028700153343379498, 0.06220453232526779, -0.04495514929294586, -0.0065127890557050705, -0.06172054633498192, -0.009365018457174301, -0.02464917115867138, -0.035406388342380524, 0.005040289834141731, -0.015868930146098137, 0.018734531477093697, 0.02240879274904728, -0.0225...
573
573
['Pascal Vincent', 'Alexandre de Brébisson', 'Xavier Bouthillier']
1412.7091v3
An important class of problems involves training deep neural networks with sparse prediction targets of very high dimension D. These occur naturally in e.g. neural language models or the learning of word-embeddings, often posed as predicting the probability of next words among a vocabulary of size D (e.g. 200 000). Com...
Efficient Exact Gradient Update for training Deep Networks with Very Large Sparse Targets
2,014
http://arxiv.org/pdf/1412.7091v3
Title Efficient Exact Gradient Update training Deep Networks Large Sparse Targets Summary important class problem involves training deep neural network sparse prediction target high dimension occur naturally eg neural language model learning wordembeddings often posed predicting probability next word among vocabulary s...
[0.02790750376880169, 0.03701736778020859, 0.024033328518271446, 0.035924483090639114, 0.025002453476190567, -0.016908086836338043, -0.024662375450134277, 0.01325901784002781, -0.06803694367408752, -0.04705901816487312, -0.037853434681892395, -0.03654506430029869, -0.007122768554836512, 0.014868839643895626, 0.02197933...
574
574
['Lili Mou', 'Hao Peng', 'Ge Li', 'Yan Xu', 'Lu Zhang', 'Zhi Jin']
1504.01106v5
This paper proposes a tree-based convolutional neural network (TBCNN) for discriminative sentence modeling. Our models leverage either constituency trees or dependency trees of sentences. The tree-based convolution process extracts sentences' structural features, and these features are aggregated by max pooling. Such a...
Discriminative Neural Sentence Modeling by Tree-Based Convolution
2,015
http://arxiv.org/pdf/1504.01106v5
Title Discriminative Neural Sentence Modeling TreeBased Convolution Summary paper proposes treebased convolutional neural network TBCNN discriminative sentence modeling model leverage either constituency tree dependency tree sentence treebased convolution process extract sentence structural feature feature aggregated m...
[0.05543934553861618, 0.02635028585791588, 0.011994539760053158, 0.09633726626634598, -0.045005373656749725, 0.015982534736394882, -0.0516539067029953, -0.03248940780758858, -0.03221478313207626, -0.07559776306152344, 1.281684603782196e-06, -0.01715259812772274, -0.030680643394589424, 0.04130610078573227, -0.0360275283...
575
575
['Han Zhao', 'Zhengdong Lu', 'Pascal Poupart']
1504.05070v2
The ability to accurately model a sentence at varying stages (e.g., word-phrase-sentence) plays a central role in natural language processing. As an effort towards this goal we propose a self-adaptive hierarchical sentence model (AdaSent). AdaSent effectively forms a hierarchy of representations from words to phrases a...
Self-Adaptive Hierarchical Sentence Model
2,015
http://arxiv.org/pdf/1504.05070v2
Title SelfAdaptive Hierarchical Sentence Model Summary ability accurately model sentence varying stage eg wordphrasesentence play central role natural language processing effort towards goal propose selfadaptive hierarchical sentence model AdaSent AdaSent effectively form hierarchy representation word phrase sentence r...
[0.05538245290517807, 0.018063362687826157, -0.032322634011507034, 0.04892224818468094, -0.05491231381893158, 0.003983187954872847, -0.0020803736988455057, -0.04065811634063721, 0.03117627464234829, -0.08946610242128372, -0.021120086312294006, 0.03724454343318939, -0.035660646855831146, 0.05444278568029404, 0.001115854...
576
576
['Cicero Nogueira dos Santos', 'Bing Xiang', 'Bowen Zhou']
1504.06580v2
Relation classification is an important semantic processing task for which state-ofthe-art systems still rely on costly handcrafted features. In this work we tackle the relation classification task using a convolutional neural network that performs classification by ranking (CR-CNN). We propose a new pairwise ranking l...
Classifying Relations by Ranking with Convolutional Neural Networks
2,015
http://arxiv.org/pdf/1504.06580v2
Title Classifying Relations Ranking Convolutional Neural Networks Summary Relation classification important semantic processing task stateoftheart system still rely costly handcrafted feature work tackle relation classification task using convolutional neural network performs classification ranking CRCNN propose new pa...
[0.07756244391202927, 0.04017649590969086, 0.007496031932532787, 0.0741201713681221, 0.000742192380130291, 0.008697119541466236, -0.0011894071940332651, 0.020560026168823242, 0.00620815996080637, -0.057921066880226135, -0.027497971430420876, 0.00218995101749897, -0.005068772938102484, -0.008564497344195843, -0.02058519...
577
577
['Thanh-Le Ha', 'Jan Niehues', 'Alex Waibel']
1504.07395v1
In this paper we combine the advantages of a model using global source sentence contexts, the Discriminative Word Lexicon, and neural networks. By using deep neural networks instead of the linear maximum entropy model in the Discriminative Word Lexicon models, we are able to leverage dependencies between different sour...
Lexical Translation Model Using a Deep Neural Network Architecture
2,015
http://arxiv.org/pdf/1504.07395v1
Title Lexical Translation Model Using Deep Neural Network Architecture Summary paper combine advantage model using global source sentence context Discriminative Word Lexicon neural network using deep neural network instead linear maximum entropy model Discriminative Word Lexicon model able leverage dependency different...
[0.04355678707361221, 0.011171996593475342, 0.007991897873580456, 0.07533075660467148, -0.052880823612213135, 0.008705146610736847, 0.010591925121843815, -0.006497151684015989, -0.04921087622642517, -0.024072065949440002, -0.04087696596980095, -0.03690309450030327, 0.04492330178618431, 0.016308343037962914, 0.029270466...
578
578
['Andrej Karpathy', 'Justin Johnson', 'Li Fei-Fei']
1506.02078v2
Recurrent Neural Networks (RNNs), and specifically a variant with Long Short-Term Memory (LSTM), are enjoying renewed interest as a result of successful applications in a wide range of machine learning problems that involve sequential data. However, while LSTMs provide exceptional results in practice, the source of the...
Visualizing and Understanding Recurrent Networks
2,015
http://arxiv.org/pdf/1506.02078v2
Title Visualizing Understanding Recurrent Networks Summary Recurrent Neural Networks RNNs specifically variant Long ShortTerm Memory LSTM enjoying renewed interest result successful application wide range machine learning problem involve sequential data However LSTMs provide exceptional result practice source performan...
[-0.011355806142091751, -0.005041065160185099, -0.01834714412689209, 0.030946750193834305, -0.02056630700826645, -0.010613009333610535, 0.018238291144371033, -0.003978530876338482, -0.042686380445957184, 0.011484013870358467, -0.0167690422385931, -0.05414002388715744, 0.057105064392089844, 0.05567161366343498, 0.024605...
579
579
['Cheng-Tao Chung', 'Cheng-Yu Tsai', 'Hsiang-Hung Lu', 'Yuan-ming Liou', 'Yen-Chen Wu', 'Yen-Ju Lu', 'Hung-yi Lee', 'Lin-shan Lee']
1506.02327v1
This paper summarizes the work done by the authors for the Zero Resource Speech Challenge organized in the technical program of Interspeech 2015. The goal of the challenge is to discover linguistic units directly from unlabeled speech data. The Multi-layered Acoustic Tokenizer (MAT) proposed in this work automatically ...
A Multi-layered Acoustic Tokenizing Deep Neural Network (MAT-DNN) for Unsupervised Discovery of Linguistic Units and Generation of High Quality Features
2,015
http://arxiv.org/pdf/1506.02327v1
Title Multilayered Acoustic Tokenizing Deep Neural Network MATDNN Unsupervised Discovery Linguistic Units Generation High Quality Features Summary paper summarizes work done author Zero Resource Speech Challenge organized technical program Interspeech 2015 goal challenge discover linguistic unit directly unlabeled spee...
[-0.023208608850836754, 0.03669661656022072, 0.011429867707192898, 0.01780596747994423, -0.01536133885383606, -0.038021013140678406, 0.07236523181200027, -0.019846657291054726, -0.10259981453418732, -0.01436596643179655, -0.0335264652967453, 0.04753900691866875, -0.011864783242344856, 0.043359726667404175, 0.0296250060...
580
580
['Douglas Bagnall']
1506.04891v2
Recurrent neural networks (RNNs) are very good at modelling the flow of text, but typically need to be trained on a far larger corpus than is available for the PAN 2015 Author Identification task. This paper describes a novel approach where the output layer of a character-level RNN language model is split into several ...
Author Identification using Multi-headed Recurrent Neural Networks
2,015
http://arxiv.org/pdf/1506.04891v2
Title Author Identification using Multiheaded Recurrent Neural Networks Summary Recurrent neural network RNNs good modelling flow text typically need trained far larger corpus available PAN 2015 Author Identification task paper describes novel approach output layer characterlevel RNN language model split several indepe...
[0.041541144251823425, 0.07551287859678268, -0.014381404966115952, 0.06047964096069336, -0.008142398670315742, 0.01562241930514574, 0.06985439360141754, -0.014277661219239235, 0.021943392232060432, 0.013253385201096535, 0.0024878140538930893, -0.008008643053472042, 0.02791939303278923, -0.02467748336493969, 0.005771238...
581
581
['Fandong Meng', 'Zhengdong Lu', 'Zhaopeng Tu', 'Hang Li', 'Qun Liu']
1506.06442v4
We propose DEEPMEMORY, a novel deep architecture for sequence-to-sequence learning, which performs the task through a series of nonlinear transformations from the representation of the input sequence (e.g., a Chinese sentence) to the final output sequence (e.g., translation to English). Inspired by the recently propose...
A Deep Memory-based Architecture for Sequence-to-Sequence Learning
2,015
http://arxiv.org/pdf/1506.06442v4
Title Deep Memorybased Architecture SequencetoSequence Learning Summary propose DEEPMEMORY novel deep architecture sequencetosequence learning performs task series nonlinear transformation representation input sequence eg Chinese sentence final output sequence eg translation English Inspired recently proposed Neural Tu...
[0.028884893283247948, 0.06040644273161888, -0.02255413681268692, 0.030304238200187683, -0.01584515906870365, 0.012662301771342754, 0.010223411954939365, -0.003853060305118561, -0.012946471571922302, -0.026064522564411163, -0.023150455206632614, -0.04285856708884239, 0.011858006939291954, 0.05385228246450424, 0.0052147...
582
582
['Ankit Kumar', 'Ozan Irsoy', 'Peter Ondruska', 'Mohit Iyyer', 'James Bradbury', 'Ishaan Gulrajani', 'Victor Zhong', 'Romain Paulus', 'Richard Socher']
1506.07285v5
Most tasks in natural language processing can be cast into question answering (QA) problems over language input. We introduce the dynamic memory network (DMN), a neural network architecture which processes input sequences and questions, forms episodic memories, and generates relevant answers. Questions trigger an itera...
Ask Me Anything: Dynamic Memory Networks for Natural Language Processing
2,015
http://arxiv.org/pdf/1506.07285v5
Title Ask Anything Dynamic Memory Networks Natural Language Processing Summary task natural language processing cast question answering QA problem language input introduce dynamic memory network DMN neural network architecture process input sequence question form episodic memory generates relevant answer Questions trig...
[0.05158541724085808, 0.02812228724360466, -0.02276724763214588, 0.02513764426112175, -0.013104015029966831, 0.012974195182323456, 0.020096927881240845, 0.005087028257548809, 0.0018260703654959798, -0.05064588785171509, 0.012586611323058605, -0.014207405969500542, -0.007919731549918652, 0.08814384788274765, 0.022306222...
583
583
['Lantian Li', 'Yiye Lin', 'Zhiyong Zhang', 'Dong Wang']
1506.08349v1
A deep learning approach has been proposed recently to derive speaker identifies (d-vector) by a deep neural network (DNN). This approach has been applied to text-dependent speaker recognition tasks and shows reasonable performance gains when combined with the conventional i-vector approach. Although promising, the exi...
Improved Deep Speaker Feature Learning for Text-Dependent Speaker Recognition
2,015
http://arxiv.org/pdf/1506.08349v1
Title Improved Deep Speaker Feature Learning TextDependent Speaker Recognition Summary deep learning approach proposed recently derive speaker identifies dvector deep neural network DNN approach applied textdependent speaker recognition task show reasonable performance gain combined conventional ivector approach Althou...
[-0.002774872351437807, 0.037153977900743484, -0.0013352383393794298, 0.021412957459688187, -0.04566184803843498, 0.02175619639456272, 0.05310129374265671, -0.01793704181909561, -0.021418150514364243, -0.01175806112587452, -0.0002018620871240273, -0.016237840056419373, 0.04732605814933777, -0.021833205595612526, -0.029...
584
584
['Nal Kalchbrenner', 'Ivo Danihelka', 'Alex Graves']
1507.01526v3
This paper introduces Grid Long Short-Term Memory, a network of LSTM cells arranged in a multidimensional grid that can be applied to vectors, sequences or higher dimensional data such as images. The network differs from existing deep LSTM architectures in that the cells are connected between network layers as well as ...
Grid Long Short-Term Memory
2,015
http://arxiv.org/pdf/1507.01526v3
Title Grid Long ShortTerm Memory Summary paper introduces Grid Long ShortTerm Memory network LSTM cell arranged multidimensional grid applied vector sequence higher dimensional data image network differs existing deep LSTM architecture cell connected network layer well along spatiotemporal dimension data network provid...
[0.0004072290612384677, 0.03275850787758827, -0.0063985129818320274, 0.07208779454231262, 0.00855559017509222, 0.025971632450819016, 0.046660978347063065, -0.023915085941553116, -0.013846147805452347, -0.008136903867125511, -0.03842177987098694, -0.10288789123296738, 0.04769172519445419, 0.04459302872419357, 0.04737400...
585
585
['Yang Liu', 'Furu Wei', 'Sujian Li', 'Heng Ji', 'Ming Zhou', 'Houfeng Wang']
1507.04646v1
Previous research on relation classification has verified the effectiveness of using dependency shortest paths or subtrees. In this paper, we further explore how to make full use of the combination of these dependency information. We first propose a new structure, termed augmented dependency path (ADP), which is compos...
A Dependency-Based Neural Network for Relation Classification
2,015
http://arxiv.org/pdf/1507.04646v1
Title DependencyBased Neural Network Relation Classification Summary Previous research relation classification verified effectiveness using dependency shortest path subtrees paper explore make full use combination dependency information first propose new structure termed augmented dependency path ADP composed shortest ...
[0.03561617434024811, 0.046365831047296524, -0.0013076739851385355, 0.07303697615861893, -0.00864286720752716, -0.01818736642599106, -0.06447212398052216, 0.028535889461636543, 0.008817428722977638, -0.04436936601996422, 0.0034797396510839462, 0.02065291255712509, 0.01121324859559536, -0.013998960144817829, -0.02868277...
586
586
['Jian Tang', 'Meng Qu', 'Qiaozhu Mei']
1508.00200v1
Unsupervised text embedding methods, such as Skip-gram and Paragraph Vector, have been attracting increasing attention due to their simplicity, scalability, and effectiveness. However, comparing to sophisticated deep learning architectures such as convolutional neural networks, these methods usually yield inferior resu...
PTE: Predictive Text Embedding through Large-scale Heterogeneous Text Networks
2,015
http://arxiv.org/pdf/1508.00200v1
Title PTE Predictive Text Embedding Largescale Heterogeneous Text Networks Summary Unsupervised text embedding method Skipgram Paragraph Vector attracting increasing attention due simplicity scalability effectiveness However comparing sophisticated deep learning architecture convolutional neural network method usually ...
[0.010130133479833603, -0.006559894885867834, 0.008196325041353703, 0.08284035325050354, -0.006328774616122246, -0.007417625281959772, 0.03629646450281143, 0.010787906125187874, 0.05100501701235771, -0.03620427846908569, -0.02727072685956955, 0.007336152717471123, -0.0077982097864151, 0.04922201484441757, 0.02580201812...
587
587
['Dongxu Zhang', 'Dong Wang']
1508.01006v2
Deep learning has gained much success in sentence-level relation classification. For example, convolutional neural networks (CNN) have delivered competitive performance without much effort on feature engineering as the conventional pattern-based methods. Thus a lot of works have been produced based on CNN structures. H...
Relation Classification via Recurrent Neural Network
2,015
http://arxiv.org/pdf/1508.01006v2
Title Relation Classification via Recurrent Neural Network Summary Deep learning gained much success sentencelevel relation classification example convolutional neural network CNN delivered competitive performance without much effort feature engineering conventional patternbased method Thus lot work produced based CNN ...
[0.06687229871749878, 0.04454890638589859, -0.002900527324527502, 0.08168329298496246, -0.041271060705184937, -0.015379820019006729, -0.004754744004458189, 0.02586425095796585, -0.014359858818352222, -0.06835654377937317, 0.007101353257894516, -0.036215901374816895, 0.01686316914856434, 0.00855633057653904, -0.03326073...
588
588
['Dongxu Zhang', 'Tianyi Luo', 'Dong Wang', 'Rong Liu']
1508.01011v1
Latent Dirichlet Allocation (LDA) is a three-level hierarchical Bayesian model for topic inference. In spite of its great success, inferring the latent topic distribution with LDA is time-consuming. Motivated by the transfer learning approach proposed by~\newcite{hinton2015distilling}, we present a novel method that us...
Learning from LDA using Deep Neural Networks
2,015
http://arxiv.org/pdf/1508.01011v1
Title Learning LDA using Deep Neural Networks Summary Latent Dirichlet Allocation LDA threelevel hierarchical Bayesian model topic inference spite great success inferring latent topic distribution LDA timeconsuming Motivated transfer learning approach proposed bynewcitehinton2015distilling present novel method us LDA s...
[0.04232034459710121, 0.05954749137163162, -0.012289010919630527, 0.027016060426831245, -0.02140469290316105, 0.014207686297595501, 0.038581497967243195, -0.012945486232638359, -0.04322564974427223, -0.044133830815553665, -0.01582692191004753, 0.017526760697364807, -0.026834798976778984, 0.030297458171844482, -0.020221...
589
589
['Marek Rei']
1508.03854v1
We investigate an extension of continuous online learning in recurrent neural network language models. The model keeps a separate vector representation of the current unit of text being processed and adaptively adjusts it after each prediction. The initial experiments give promising results, indicating that the method ...
Online Representation Learning in Recurrent Neural Language Models
2,015
http://arxiv.org/pdf/1508.03854v1
Title Online Representation Learning Recurrent Neural Language Models Summary investigate extension continuous online learning recurrent neural network language model model keep separate vector representation current unit text processed adaptively adjusts prediction initial experiment give promising result indicating m...
[0.03535664826631546, -0.001378671615384519, -0.01852804608643055, 0.04573841765522957, -0.020446332171559334, -0.0014789568958804011, 0.01775195635855198, 0.017491959035396576, -0.0062713478691875935, -0.0704064890742302, -0.0024637607857584953, -0.052236977964639664, 0.029771896079182625, 0.07973246276378632, 0.00975...
590
590
['Ye Zhang', 'Byron Wallace']
1510.03820v4
Convolutional Neural Networks (CNNs) have recently achieved remarkably strong performance on the practically important task of sentence classification (kim 2014, kalchbrenner 2014, johnson 2014). However, these models require practitioners to specify an exact model architecture and set accompanying hyperparameters, inc...
A Sensitivity Analysis of (and Practitioners' Guide to) Convolutional Neural Networks for Sentence Classification
2,015
http://arxiv.org/pdf/1510.03820v4
Title Sensitivity Analysis Practitioners Guide Convolutional Neural Networks Sentence Classification Summary Convolutional Neural Networks CNNs recently achieved remarkably strong performance practically important task sentence classification kim 2014 kalchbrenner 2014 johnson 2014 However model require practitioner sp...
[0.07863238453865051, 0.04088165611028671, -0.023126494139432907, 0.05393735319375992, -0.03360119089484215, 0.0050298250280320644, 0.028157170861959457, 0.02379668690264225, -0.020739585161209106, -0.06305491179227829, -0.00040149083361029625, -0.00851241685450077, -0.009140674024820328, 0.011043904349207878, -0.00964...
591
591
['Yu Zhang', 'Ekapol Chuangsuwanich', 'James Glass', 'Dong Yu']
1510.08985v1
In this paper, we investigate the use of prediction-adaptation-correction recurrent neural networks (PAC-RNNs) for low-resource speech recognition. A PAC-RNN is comprised of a pair of neural networks in which a {\it correction} network uses auxiliary information given by a {\it prediction} network to help estimate the ...
Prediction-Adaptation-Correction Recurrent Neural Networks for Low-Resource Language Speech Recognition
2,015
http://arxiv.org/pdf/1510.08985v1
Title PredictionAdaptationCorrection Recurrent Neural Networks LowResource Language Speech Recognition Summary paper investigate use predictionadaptationcorrection recurrent neural network PACRNNs lowresource speech recognition PACRNN comprised pair neural network correction network us auxiliary information given predi...
[0.00022266522864811122, 0.014842996373772621, 0.004538939334452152, 0.047181304544210434, -0.009879344142973423, -0.025139303877949715, 0.005989015102386475, -0.003311931388452649, -0.02159503474831581, -0.01609487272799015, -0.03978554904460907, -0.039478711783885956, 0.0883834958076477, 0.03858489543199539, 0.007793...
592
592
['Hongyu Guo']
1510.09202v1
We introduce a novel schema for sequence to sequence learning with a Deep Q-Network (DQN), which decodes the output sequence iteratively. The aim here is to enable the decoder to first tackle easier portions of the sequences, and then turn to cope with difficult parts. Specifically, in each iteration, an encoder-decode...
Generating Text with Deep Reinforcement Learning
2,015
http://arxiv.org/pdf/1510.09202v1
Title Generating Text Deep Reinforcement Learning Summary introduce novel schema sequence sequence learning Deep QNetwork DQN decodes output sequence iteratively aim enable decoder first tackle easier portion sequence turn cope difficult part Specifically iteration encoderdecoder Long ShortTerm Memory LSTM network empl...
[0.050332311540842056, 0.03927690535783768, 0.002629154361784458, 0.012904432602226734, -0.016218572854995728, -0.004598155617713928, -0.018037695437669754, 0.0040520173497498035, -0.0072245425544679165, -0.015038839541375637, 0.003156442893669009, 0.005189321935176849, -0.0173582062125206, 0.06151720881462097, 0.00103...
593
593
['Junyoung Chung', 'Jacob Devlin', 'Hany Hassan Awadalla']
1511.01042v2
In this paper, we explore different neural network architectures that can predict if a speaker of a given utterance is asking a question or making a statement. We com- pare the outcomes of regularization methods that are popularly used to train deep neural networks and study how different context functions can affect t...
Detecting Interrogative Utterances with Recurrent Neural Networks
2,015
http://arxiv.org/pdf/1511.01042v2
Title Detecting Interrogative Utterances Recurrent Neural Networks Summary paper explore different neural network architecture predict speaker given utterance asking question making statement com pare outcome regularization method popularly used train deep neural network study different context function affect classifi...
[0.07140016555786133, 0.028240378946065903, -0.0030725793913006783, 0.05219525471329689, -0.05397335812449455, -0.021107256412506104, 0.016506560146808624, 0.009133312851190567, 0.00037390904617495835, -0.0003898455179296434, -0.0007419760804623365, -0.021651597693562508, 0.00403172941878438, 0.03598712384700775, 0.008...
594
594
['Navdeep Jaitly', 'David Sussillo', 'Quoc V. Le', 'Oriol Vinyals', 'Ilya Sutskever', 'Samy Bengio']
1511.04868v4
Sequence-to-sequence models have achieved impressive results on various tasks. However, they are unsuitable for tasks that require incremental predictions to be made as more data arrives or tasks that have long input sequences and output sequences. This is because they generate an output sequence conditioned on an enti...
A Neural Transducer
2,015
http://arxiv.org/pdf/1511.04868v4
Title Neural Transducer Summary Sequencetosequence model achieved impressive result various task However unsuitable task require incremental prediction made data arrives task long input sequence output sequence generate output sequence conditioned entire input sequence paper present Neural Transducer make incremental p...
[0.015562239103019238, 0.054453905671834946, -0.0022053758148103952, -0.013406692072749138, -0.014066972769796848, -0.031723491847515106, 0.003930010832846165, 0.01667247898876667, -0.050774432718753815, -0.0017024906119331717, 0.039740875363349915, 0.01903730072081089, 0.021850761026144028, 0.0713837593793869, 0.03639...
595
595
['Ethan Caballero']
1511.06420v2
Question Answering (QA) is fundamental to natural language processing in that most nlp problems can be phrased as QA (Kumar et al., 2015). Current weakly supervised memory network models that have been proposed so far struggle at answering questions that involve relations among multiple entities (such as facebook's bAb...
Skip-Thought Memory Networks
2,015
http://arxiv.org/pdf/1511.06420v2
Title SkipThought Memory Networks Summary Question Answering QA fundamental natural language processing nlp problem phrased QA Kumar et al 2015 Current weakly supervised memory network model proposed far struggle answering question involve relation among multiple entity facebooks bAbi qa5threeargrelations Weston et al ...
[0.04113145172595978, 0.017919592559337616, -0.005954056512564421, 0.06399160623550415, -0.021321507170796394, 0.032747529447078705, 0.019004343077540398, -0.027221692726016045, 0.02152428589761257, -0.05416605994105339, -0.015545212663710117, -0.015252524055540562, -0.0282201636582613, 0.023438099771738052, 0.03672714...
596
596
['Jason P. C. Chiu', 'Eric Nichols']
1511.08308v5
Named entity recognition is a challenging task that has traditionally required large amounts of knowledge in the form of feature engineering and lexicons to achieve high performance. In this paper, we present a novel neural network architecture that automatically detects word- and character-level features using a hybri...
Named Entity Recognition with Bidirectional LSTM-CNNs
2,015
http://arxiv.org/pdf/1511.08308v5
Title Named Entity Recognition Bidirectional LSTMCNNs Summary Named entity recognition challenging task traditionally required large amount knowledge form feature engineering lexicon achieve high performance paper present novel neural network architecture automatically detects word characterlevel feature using hybrid b...
[0.024691695347428322, 0.0367535836994648, 0.033279772847890854, 0.09188473224639893, -0.01054808497428894, 0.02476241998374462, 0.010465210303664207, 0.051607243716716766, 0.016980402171611786, -0.05492996796965599, -0.009933359920978546, -0.03147837892174721, -0.02064288593828678, 0.055758364498615265, 0.032964557409...
597
597
['Konstantin Lopyrev']
1512.01712v1
We describe an application of an encoder-decoder recurrent neural network with LSTM units and attention to generating headlines from the text of news articles. We find that the model is quite effective at concisely paraphrasing news articles. Furthermore, we study how the neural network decides which input words to pay...
Generating News Headlines with Recurrent Neural Networks
2,015
http://arxiv.org/pdf/1512.01712v1
Title Generating News Headlines Recurrent Neural Networks Summary describe application encoderdecoder recurrent neural network LSTM unit attention generating headline text news article find model quite effective concisely paraphrasing news article Furthermore study neural network decides input word pay attention specif...
[0.049424909055233, 0.015581775456666946, 0.013913068920373917, 0.029738688841462135, -0.02919752337038517, 0.0008220571326091886, 0.017116209492087364, 0.009740505367517471, -0.027730794623494148, -0.03231985494494438, 0.014293843880295753, 0.030253801494836807, -0.0017017943318933249, 0.049015358090400696, 0.01676741...
598
598
['Pranjal Singh', 'Amitabha Mukerjee']
1512.03549v1
Despite the success of distributional semantics, composing phrases from word vectors remains an important challenge. Several methods have been tried for benchmark tasks such as sentiment classification, including word vector averaging, matrix-vector approaches based on parsing, and on-the-fly learning of paragraph vect...
Words are not Equal: Graded Weighting Model for building Composite Document Vectors
2,015
http://arxiv.org/pdf/1512.03549v1
Title Words Equal Graded Weighting Model building Composite Document Vectors Summary Despite success distributional semantics composing phrase word vector remains important challenge Several method tried benchmark task sentiment classification including word vector averaging matrixvector approach based parsing onthefly...
[0.08187291771173477, 0.06270714104175568, 0.017307894304394722, 0.03457587584853172, -0.05003199353814125, 0.01512089278548956, 0.007526844274252653, 0.031498175114393234, -0.04426541179418564, -0.07682067900896072, -0.004928107373416424, -0.02226550504565239, 0.03401249274611473, 0.014045490883290768, -0.023086180910...
599
599
['Liang Lu', 'Steve Renals']
1512.04280v4
For speech recognition, deep neural networks (DNNs) have significantly improved the recognition accuracy in most of benchmark datasets and application domains. However, compared to the conventional Gaussian mixture models, DNN-based acoustic models usually have much larger number of model parameters, making it challeng...
Small-footprint Deep Neural Networks with Highway Connections for Speech Recognition
2,015
http://arxiv.org/pdf/1512.04280v4
Title Smallfootprint Deep Neural Networks Highway Connections Speech Recognition Summary speech recognition deep neural network DNNs significantly improved recognition accuracy benchmark datasets application domain However compared conventional Gaussian mixture model DNNbased acoustic model usually much larger number m...
[-0.02974539063870907, 0.05479646474123001, -0.020276816561818123, 0.05071445554494858, -0.0004927068948745728, -0.015778936445713043, 0.05087105557322502, -0.005609317682683468, -0.06028558313846588, 0.004958346951752901, -0.001501046703197062, -0.013900922611355782, 0.052471064031124115, 0.02007078193128109, 0.017005...