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
1,800
1,800
['Nana Li', 'Shuangfei Zhai', 'Zhongfei Zhang', 'Boying Liu']
1611.08737v1
Structural correspondence learning (SCL) is an effective method for cross-lingual sentiment classification. This approach uses unlabeled documents along with a word translation oracle to automatically induce task specific, cross-lingual correspondences. It transfers knowledge through identifying important features, i.e...
Structural Correspondence Learning for Cross-lingual Sentiment Classification with One-to-many Mappings
2,016
http://arxiv.org/pdf/1611.08737v1
Title Structural Correspondence Learning Crosslingual Sentiment Classification Onetomany Mappings Summary Structural correspondence learning SCL effective method crosslingual sentiment classification approach us unlabeled document along word translation oracle automatically induce task specific crosslingual corresponde...
[0.0012674559839069843, 0.03630300983786583, 0.011639495380222797, 0.04314139485359192, -0.05940135940909386, 0.022400718182325363, 0.002370640402659774, -0.008014705963432789, 0.01659672148525715, -0.045232247561216354, -0.054853588342666626, 0.009036398492753506, 0.023090790957212448, 0.01853318139910698, -0.03270522...
1,801
1,801
['Arvind Neelakantan', 'Quoc V. Le', 'Martin Abadi', 'Andrew McCallum', 'Dario Amodei']
1611.08945v4
Learning a natural language interface for database tables is a challenging task that involves deep language understanding and multi-step reasoning. The task is often approached by mapping natural language queries to logical forms or programs that provide the desired response when executed on the database. To our knowle...
Learning a Natural Language Interface with Neural Programmer
2,016
http://arxiv.org/pdf/1611.08945v4
Title Learning Natural Language Interface Neural Programmer Summary Learning natural language interface database table challenging task involves deep language understanding multistep reasoning task often approached mapping natural language query logical form program provide desired response executed database knowledge ...
[0.05552467331290245, 0.03418844938278198, -0.001780361752025783, 0.03933074697852135, -0.020683733746409416, 0.025561412796378136, -0.01044963113963604, -0.0325695164501667, 0.030906934291124344, -0.04440595582127571, 0.0024919600691646338, 0.032392654567956924, -0.02605617418885231, 0.050485339015722275, 0.0171355288...
1,802
1,802
['Brian Patton', 'Yannis Agiomyrgiannakis', 'Michael Terry', 'Kevin Wilson', 'Rif A. Saurous', 'D. Sculley']
1611.09207v1
Developers of text-to-speech synthesizers (TTS) often make use of human raters to assess the quality of synthesized speech. We demonstrate that we can model human raters' mean opinion scores (MOS) of synthesized speech using a deep recurrent neural network whose inputs consist solely of a raw waveform. Our best models ...
AutoMOS: Learning a non-intrusive assessor of naturalness-of-speech
2,016
http://arxiv.org/pdf/1611.09207v1
Title AutoMOS Learning nonintrusive assessor naturalnessofspeech Summary Developers texttospeech synthesizer TTS often make use human raters ass quality synthesized speech demonstrate model human raters mean opinion score MOS synthesized speech using deep recurrent neural network whose input consist solely raw waveform...
[-0.013415633700788021, 0.028169682249426842, 0.005776133853942156, -0.03177510201931, 0.01619931496679783, -0.05052752047777176, 0.016736896708607674, -0.020463120192289352, -0.03631516918540001, -0.002587981056421995, -0.09111268818378448, -0.01690007746219635, 0.036594316363334656, 0.07706080377101898, 0.03557543829...
1,803
1,803
['Jian Tang', 'Meng Qu', 'Qiaozhu Mei']
1611.09878v1
Most existing word embedding approaches do not distinguish the same words in different contexts, therefore ignoring their contextual meanings. As a result, the learned embeddings of these words are usually a mixture of multiple meanings. In this paper, we acknowledge multiple identities of the same word in different co...
Identity-sensitive Word Embedding through Heterogeneous Networks
2,016
http://arxiv.org/pdf/1611.09878v1
Title Identitysensitive Word Embedding Heterogeneous Networks Summary existing word embedding approach distinguish word different context therefore ignoring contextual meaning result learned embeddings word usually mixture multiple meaning paper acknowledge multiple identity word different context learn textbfidentitys...
[0.017564360052347183, 0.016554193571209908, 0.004579450935125351, 0.07197198271751404, -0.0401650033891201, 0.02678823657333851, 0.0058804950676858425, -0.01859535276889801, 0.005805789493024349, -0.06446308642625809, -0.049478452652692795, -0.007628738414496183, 0.0019838621374219656, 0.0279469545930624, 0.0104464702...
1,804
1,804
['Xuezhe Ma', 'Eduard Hovy']
1701.00874v4
In this paper, we propose a probabilistic parsing model, which defines a proper conditional probability distribution over non-projective dependency trees for a given sentence, using neural representations as inputs. The neural network architecture is based on bi-directional LSTM-CNNs which benefits from both word- and ...
Neural Probabilistic Model for Non-projective MST Parsing
2,017
http://arxiv.org/pdf/1701.00874v4
Title Neural Probabilistic Model Nonprojective MST Parsing Summary paper propose probabilistic parsing model defines proper conditional probability distribution nonprojective dependency tree given sentence using neural representation input neural network architecture based bidirectional LSTMCNNs benefit word characterl...
[0.0291071068495512, 0.05357038602232933, 0.007234678603708744, 0.05150732398033142, -0.04941926151514053, 0.0037337830290198326, -0.024954041466116905, -0.045751381665468216, -0.01896490715444088, -0.04684773460030556, 0.022243570536375046, -0.021301323547959328, -0.0003236742049921304, 0.065390944480896, 0.0175466369...
1,805
1,805
['Ying Zhang', 'Mohammad Pezeshki', 'Philemon Brakel', 'Saizheng Zhang', 'Cesar Laurent Yoshua Bengio', 'Aaron Courville']
1701.02720v1
Convolutional Neural Networks (CNNs) are effective models for reducing spectral variations and modeling spectral correlations in acoustic features for automatic speech recognition (ASR). Hybrid speech recognition systems incorporating CNNs with Hidden Markov Models/Gaussian Mixture Models (HMMs/GMMs) have achieved the ...
Towards End-to-End Speech Recognition with Deep Convolutional Neural Networks
2,017
http://arxiv.org/pdf/1701.02720v1
Title Towards EndtoEnd Speech Recognition Deep Convolutional Neural Networks Summary Convolutional Neural Networks CNNs effective model reducing spectral variation modeling spectral correlation acoustic feature automatic speech recognition ASR Hybrid speech recognition system incorporating CNNs Hidden Markov ModelsGaus...
[0.017318621277809143, 0.02971363440155983, 0.03208262473344803, 0.02335193008184433, 0.007083885837346315, -0.00831450056284666, 0.024522332474589348, -0.022249514237046242, -0.01867346279323101, 0.013924059458076954, -0.041971102356910706, -0.017230898141860962, 0.03422018513083458, 0.038362447172403336, -0.005477155...
1,806
1,806
['Avner May', 'Alireza Bagheri Garakani', 'Zhiyun Lu', 'Dong Guo', 'Kuan Liu', 'Aurélien Bellet', 'Linxi Fan', 'Michael Collins', 'Daniel Hsu', 'Brian Kingsbury', 'Michael Picheny', 'Fei Sha']
1701.03577v1
We study large-scale kernel methods for acoustic modeling in speech recognition and compare their performance to deep neural networks (DNNs). We perform experiments on four speech recognition datasets, including the TIMIT and Broadcast News benchmark tasks, and compare these two types of models on frame-level performan...
Kernel Approximation Methods for Speech Recognition
2,017
http://arxiv.org/pdf/1701.03577v1
Title Kernel Approximation Methods Speech Recognition Summary study largescale kernel method acoustic modeling speech recognition compare performance deep neural network DNNs perform experiment four speech recognition datasets including TIMIT Broadcast News benchmark task compare two type model framelevel performance m...
[-0.007406337186694145, 0.02502695471048355, 0.00043601091601885855, 0.08287081867456436, -0.009637562558054924, -0.030430292710661888, 0.024150796234607697, 0.031682513654232025, -0.008536962792277336, -0.00719692325219512, -0.07016468793153763, 0.008084210567176342, 0.04014984890818596, 0.03296871855854988, -0.027503...
1,807
1,807
['Patrick Ng']
1701.06279v1
One of the ubiquitous representation of long DNA sequence is dividing it into shorter k-mer components. Unfortunately, the straightforward vector encoding of k-mer as a one-hot vector is vulnerable to the curse of dimensionality. Worse yet, the distance between any pair of one-hot vectors is equidistant. This is partic...
dna2vec: Consistent vector representations of variable-length k-mers
2,017
http://arxiv.org/pdf/1701.06279v1
Title dna2vec Consistent vector representation variablelength kmers Summary One ubiquitous representation long DNA sequence dividing shorter kmer component Unfortunately straightforward vector encoding kmer onehot vector vulnerable curse dimensionality Worse yet distance pair onehot vector equidistant particularly prob...
[0.036550890654325485, 0.03986657038331032, -0.016594476997852325, -0.007526705507189035, -0.02098468504846096, 0.010443637147545815, 0.03193220868706703, 0.019121946766972542, -0.03656378388404846, -0.006672477349638939, 0.055839117616415024, -0.023255163803696632, 0.06256533414125443, 0.03052331693470478, 0.052499640...
1,808
1,808
['Vinci Chow']
1701.08711v3
In Chinese societies, superstition is of paramount importance, and vehicle license plates with desirable numbers can fetch very high prices in auctions. Unlike other valuable items, license plates are not allocated an estimated price before auction. I propose that the task of predicting plate prices can be viewed as a ...
Predicting Auction Price of Vehicle License Plate with Deep Recurrent Neural Network
2,017
http://arxiv.org/pdf/1701.08711v3
Title Predicting Auction Price Vehicle License Plate Deep Recurrent Neural Network Summary Chinese society superstition paramount importance vehicle license plate desirable number fetch high price auction Unlike valuable item license plate allocated estimated price auction propose task predicting plate price viewed nat...
[0.03520694747567177, 0.0917019322514534, -0.001388432807289064, 0.06181337684392929, -0.007320952136069536, 0.0014056572690606117, 0.031645845621824265, 0.03662719950079918, -0.023317934945225716, -0.032684795558452606, 0.004996344447135925, -0.022363590076565742, 0.018301939591765404, 0.060761816799640656, 0.01563965...
1,809
1,809
['Alan Mishler', 'Kevin Wonus', 'Wendy Chambers', 'Michael Bloodgood']
1702.06216v2
Since the events of the Arab Spring, there has been increased interest in using social media to anticipate social unrest. While efforts have been made toward automated unrest prediction, we focus on filtering the vast volume of tweets to identify tweets relevant to unrest, which can be provided to downstream users for ...
Filtering Tweets for Social Unrest
2,017
http://arxiv.org/pdf/1702.06216v2
Title Filtering Tweets Social Unrest Summary Since event Arab Spring increased interest using social medium anticipate social unrest effort made toward automated unrest prediction focus filtering vast volume tweet identify tweet relevant unrest provided downstream user analysis train supervised classifier able label Ar...
[0.034279439598321915, 0.04583505168557167, -0.005827355198562145, -0.040409285575151443, -0.011834709905087948, 0.020255493000149727, 0.05957356467843056, -0.007023077458143234, 0.06593460589647293, -0.03307213634252548, 0.07932148873806, -0.0059230737388134, -0.05687170848250389, 0.0461648590862751, -0.02236612699925...
1,810
1,810
['Mark Belford', 'Brian Mac Namee', 'Derek Greene']
1702.07186v2
Topic models can provide us with an insight into the underlying latent structure of a large corpus of documents. A range of methods have been proposed in the literature, including probabilistic topic models and techniques based on matrix factorization. However, in both cases, standard implementations rely on stochastic...
Stability of Topic Modeling via Matrix Factorization
2,017
http://arxiv.org/pdf/1702.07186v2
Title Stability Topic Modeling via Matrix Factorization Summary Topic model provide u insight underlying latent structure large corpus document range method proposed literature including probabilistic topic model technique based matrix factorization However case standard implementation rely stochastic element initializ...
[0.03557785972952843, 0.006266177631914616, -0.03552798181772232, 0.007888085208833218, -0.002053199103102088, 0.01679142378270626, -0.021947331726551056, -0.000825749069917947, -0.05312570184469223, -0.04838170111179352, 0.030046356841921806, -0.03394250199198723, 0.008130177855491638, 0.031151117756962776, -0.0426447...
1,811
1,811
['Lidong Bing', 'William W. Cohen', 'Bhuwan Dhingra']
1703.01557v2
We propose a general approach to modeling semi-supervised learning (SSL) algorithms. Specifically, we present a declarative language for modeling both traditional supervised classification tasks and many SSL heuristics, including both well-known heuristics such as co-training and novel domain-specific heuristics. In ad...
Using Graphs of Classifiers to Impose Declarative Constraints on Semi-supervised Learning
2,017
http://arxiv.org/pdf/1703.01557v2
Title Using Graphs Classifiers Impose Declarative Constraints Semisupervised Learning Summary propose general approach modeling semisupervised learning SSL algorithm Specifically present declarative language modeling traditional supervised classification task many SSL heuristic including wellknown heuristic cotraining ...
[0.04919423162937164, 0.019146768376231194, 0.005755882244557142, 0.02095760591328144, -0.07231153547763824, -0.004520369693636894, 0.03727061301469803, 0.020791033282876015, 0.06431078910827637, -0.07595660537481308, -0.008158226497471333, -0.010802973061800003, -0.029065614566206932, 0.039896637201309204, -0.03057435...
1,812
1,812
['Graham Neubig']
1703.01619v1
This tutorial introduces a new and powerful set of techniques variously called "neural machine translation" or "neural sequence-to-sequence models". These techniques have been used in a number of tasks regarding the handling of human language, and can be a powerful tool in the toolbox of anyone who wants to model seque...
Neural Machine Translation and Sequence-to-sequence Models: A Tutorial
2,017
http://arxiv.org/pdf/1703.01619v1
Title Neural Machine Translation Sequencetosequence Models Tutorial Summary tutorial introduces new powerful set technique variously called neural machine translation neural sequencetosequence model technique used number task regarding handling human language powerful tool toolbox anyone want model sequential data sort...
[0.02718224562704563, 0.07139661908149719, -0.03400327265262604, 0.04518765211105347, -0.035650674253702164, 0.021642981097102165, -0.000479767972137779, -0.006931219715625048, -0.017347078770399094, -0.008687418885529041, 0.0518849678337574, -0.06653830409049988, 0.05847102776169777, 0.04139387980103493, 0.02133645676...
1,813
1,813
['Dani Yogatama', 'Chris Dyer', 'Wang Ling', 'Phil Blunsom']
1703.01898v2
We empirically characterize the performance of discriminative and generative LSTM models for text classification. We find that although RNN-based generative models are more powerful than their bag-of-words ancestors (e.g., they account for conditional dependencies across words in a document), they have higher asymptoti...
Generative and Discriminative Text Classification with Recurrent Neural Networks
2,017
http://arxiv.org/pdf/1703.01898v2
Title Generative Discriminative Text Classification Recurrent Neural Networks Summary empirically characterize performance discriminative generative LSTM model text classification find although RNNbased generative model powerful bagofwords ancestor eg account conditional dependency across word document higher asymptoti...
[0.04899232089519501, 0.02706983871757984, 0.004009331110864878, 0.05531726032495499, -0.02502382919192314, 0.01023771706968546, 0.0022498113103210926, 0.03049919568002224, -0.004415981471538544, -0.06179029121994972, 0.0074777789413928986, -0.01659322716295719, -0.0014515104703605175, 0.017110653221607208, -0.01246130...
1,814
1,814
['Ron J. Weiss', 'Jan Chorowski', 'Navdeep Jaitly', 'Yonghui Wu', 'Zhifeng Chen']
1703.08581v2
We present a recurrent encoder-decoder deep neural network architecture that directly translates speech in one language into text in another. The model does not explicitly transcribe the speech into text in the source language, nor does it require supervision from the ground truth source language transcription during t...
Sequence-to-Sequence Models Can Directly Translate Foreign Speech
2,017
http://arxiv.org/pdf/1703.08581v2
Title SequencetoSequence Models Directly Translate Foreign Speech Summary present recurrent encoderdecoder deep neural network architecture directly translates speech one language text another model explicitly transcribe speech text source language require supervision ground truth source language transcription training...
[0.015931488946080208, 0.08791428059339523, -0.011396514251828194, 0.050817232578992844, -0.010319534689188004, -0.008381184190511703, 0.049350280314683914, -0.03355008363723755, -0.04053041711449623, 0.005900312680751085, -0.07775794714689255, -0.03215538710355759, 0.055748842656612396, 0.037915006279945374, 0.0237506...
1,815
1,815
['Eric Bailey', 'Shuchin Aeron']
1704.02686v2
Most popular word embedding techniques involve implicit or explicit factorization of a word co-occurrence based matrix into low rank factors. In this paper, we aim to generalize this trend by using numerical methods to factor higher-order word co-occurrence based arrays, or \textit{tensors}. We present four word embedd...
Word Embeddings via Tensor Factorization
2,017
http://arxiv.org/pdf/1704.02686v2
Title Word Embeddings via Tensor Factorization Summary popular word embedding technique involve implicit explicit factorization word cooccurrence based matrix low rank factor paper aim generalize trend using numerical method factor higherorder word cooccurrence based array textittensors present four word embeddings usi...
[-0.013408784754574299, 0.012375653721392155, 0.002670424757525325, 0.05957482010126114, -0.0012145967921242118, 0.02977621741592884, 0.015402008779346943, 0.023599909618496895, 0.012097876518964767, -0.011041215620934963, -0.003193911397829652, -0.002469652099534869, 0.016663141548633575, 0.01639997959136963, 0.001746...
1,816
1,816
['Zahra Mousavi', 'Heshaam Faili']
1704.03223v1
This paper presents an automated supervised method for Persian wordnet construction. Using a Persian corpus and a bi-lingual dictionary, the initial links between Persian words and Princeton WordNet synsets have been generated. These links will be discriminated later as correct or incorrect by employing seven features ...
Persian Wordnet Construction using Supervised Learning
2,017
http://arxiv.org/pdf/1704.03223v1
Title Persian Wordnet Construction using Supervised Learning Summary paper present automated supervised method Persian wordnet construction Using Persian corpus bilingual dictionary initial link Persian word Princeton WordNet synset generated link discriminated later correct incorrect employing seven feature trained cl...
[0.006227482575923204, 0.017573731020092964, -0.02447565272450447, 0.057863783091306686, -0.04580408334732056, -0.015208335593342781, 0.034283556044101715, 0.029870247468352318, 0.0277931559830904, -0.033491551876068115, -0.009829728864133358, -0.030263060703873634, 0.041625987738370895, -0.03026626817882061, 0.0223998...
1,817
1,817
['Wei-Ning Hsu', 'Yu Zhang', 'James Glass']
1704.04222v2
An ability to model a generative process and learn a latent representation for speech in an unsupervised fashion will be crucial to process vast quantities of unlabelled speech data. Recently, deep probabilistic generative models such as Variational Autoencoders (VAEs) have achieved tremendous success in modeling natur...
Learning Latent Representations for Speech Generation and Transformation
2,017
http://arxiv.org/pdf/1704.04222v2
Title Learning Latent Representations Speech Generation Transformation Summary ability model generative process learn latent representation speech unsupervised fashion crucial process vast quantity unlabelled speech data Recently deep probabilistic generative model Variational Autoencoders VAEs achieved tremendous succ...
[0.0031485827639698982, 0.0894872173666954, -0.011330259963870049, -0.0032449644058942795, -0.003966578748077154, -0.023981081321835518, 0.053813714534044266, -0.050330013036727905, -0.0973578616976738, 0.007771856151521206, -0.03999333456158638, -0.024325808510184288, 0.02355148084461689, 0.08025045692920685, 0.049356...
1,818
1,818
['Sebastien Jean', 'Stanislas Lauly', 'Orhan Firat', 'Kyunghyun Cho']
1704.05135v1
We propose a neural machine translation architecture that models the surrounding text in addition to the source sentence. These models lead to better performance, both in terms of general translation quality and pronoun prediction, when trained on small corpora, although this improvement largely disappears when trained...
Does Neural Machine Translation Benefit from Larger Context?
2,017
http://arxiv.org/pdf/1704.05135v1
Title Neural Machine Translation Benefit Larger Context Summary propose neural machine translation architecture model surrounding text addition source sentence model lead better performance term general translation quality pronoun prediction trained small corpus although improvement largely disappears trained larger co...
[0.08857234567403793, 0.05130188912153244, -0.007489745505154133, 0.0352918915450573, 0.010534964501857758, -0.0016481010243296623, 0.012680089101195335, -0.05986481532454491, -0.018240440636873245, -0.02833694964647293, -0.007832972332835197, -0.058716949075460434, 0.03813064098358154, -0.005747844465076923, 0.0654757...
1,819
1,819
['Julia Kreutzer', 'Artem Sokolov', 'Stefan Riezler']
1704.06497v1
Bandit structured prediction describes a stochastic optimization framework where learning is performed from partial feedback. This feedback is received in the form of a task loss evaluation to a predicted output structure, without having access to gold standard structures. We advance this framework by lifting linear ba...
Bandit Structured Prediction for Neural Sequence-to-Sequence Learning
2,017
http://arxiv.org/pdf/1704.06497v1
Title Bandit Structured Prediction Neural SequencetoSequence Learning Summary Bandit structured prediction describes stochastic optimization framework learning performed partial feedback feedback received form task loss evaluation predicted output structure without access gold standard structure advance framework lifti...
[0.03230813145637512, 0.06323099881410599, -0.005218170117586851, -0.0017516420921310782, 0.021685946732759476, -0.026574542745947838, 0.0032869784627109766, 0.010635502636432648, 0.007438377011567354, -0.0165310837328434, -0.02728174440562725, -0.04555603116750717, 0.009015054441988468, 0.07751011848449707, 0.01090968...
1,820
1,820
['Lijun Wu', 'Yingce Xia', 'Li Zhao', 'Fei Tian', 'Tao Qin', 'Jianhuang Lai', 'Tie-Yan Liu']
1704.06933v3
In this paper, we study a new learning paradigm for Neural Machine Translation (NMT). Instead of maximizing the likelihood of the human translation as in previous works, we minimize the distinction between human translation and the translation given by an NMT model. To achieve this goal, inspired by the recent success ...
Adversarial Neural Machine Translation
2,017
http://arxiv.org/pdf/1704.06933v3
Title Adversarial Neural Machine Translation Summary paper study new learning paradigm Neural Machine Translation NMT Instead maximizing likelihood human translation previous work minimize distinction human translation translation given NMT model achieve goal inspired recent success generative adversarial network GANs ...
[0.04822970926761627, 0.04422661289572716, -0.013169092126190662, 0.06206069886684418, -0.03406056761741638, -0.007599533535540104, 0.02515542134642601, 0.022461971268057823, -0.021381737664341927, -0.029070058837532997, -0.03283661603927612, -0.03485105186700821, 0.02278311364352703, -0.029760409146547318, 0.074840173...
1,821
1,821
['Michael Bloodgood', 'Benjamin Strauss']
1704.07050v2
Global constraints and reranking have not been used in cognates detection research to date. We propose methods for using global constraints by performing rescoring of the score matrices produced by state of the art cognates detection systems. Using global constraints to perform rescoring is complementary to state of th...
Using Global Constraints and Reranking to Improve Cognates Detection
2,017
http://arxiv.org/pdf/1704.07050v2
Title Using Global Constraints Reranking Improve Cognates Detection Summary Global constraint reranking used cognate detection research date propose method using global constraint performing rescoring score matrix produced state art cognate detection system Using global constraint perform rescoring complementary state ...
[0.04228793457150459, 0.06878424435853958, -0.04401877149939537, 0.06750334799289703, -0.05544128641486168, 0.029519097879529, 0.009530557319521904, 0.03272726759314537, 0.02947760932147503, -0.028111495077610016, -0.015730177983641624, -0.01048736646771431, 0.028951693326234818, 0.01066756434738636, -0.018563058227300...
1,822
1,822
['Jonathan Chang', 'Stefan Scherer']
1705.02394v1
Automatically assessing emotional valence in human speech has historically been a difficult task for machine learning algorithms. The subtle changes in the voice of the speaker that are indicative of positive or negative emotional states are often "overshadowed" by voice characteristics relating to emotional intensity ...
Learning Representations of Emotional Speech with Deep Convolutional Generative Adversarial Networks
2,017
http://arxiv.org/pdf/1705.02394v1
Title Learning Representations Emotional Speech Deep Convolutional Generative Adversarial Networks Summary Automatically assessing emotional valence human speech historically difficult task machine learning algorithm subtle change voice speaker indicative positive negative emotional state often overshadowed voice chara...
[0.01928901858627796, 0.07107654213905334, -0.015316963195800781, -0.0023292917758226395, 0.02279370278120041, 0.011128722690045834, 0.04492994025349617, -0.023671584203839302, 0.009613116271793842, 0.007953130640089512, -0.09080372005701065, -0.020096665248274803, -0.020327309146523476, 0.03376932442188263, 0.00234634...
1,823
1,823
['Ming Sun', 'Anirudh Raju', 'George Tucker', 'Sankaran Panchapagesan', 'Gengshen Fu', 'Arindam Mandal', 'Spyros Matsoukas', 'Nikko Strom', 'Shiv Vitaladevuni']
1705.02411v1
We propose a max-pooling based loss function for training Long Short-Term Memory (LSTM) networks for small-footprint keyword spotting (KWS), with low CPU, memory, and latency requirements. The max-pooling loss training can be further guided by initializing with a cross-entropy loss trained network. A posterior smoothin...
Max-Pooling Loss Training of Long Short-Term Memory Networks for Small-Footprint Keyword Spotting
2,017
http://arxiv.org/pdf/1705.02411v1
Title MaxPooling Loss Training Long ShortTerm Memory Networks SmallFootprint Keyword Spotting Summary propose maxpooling based loss function training Long ShortTerm Memory LSTM network smallfootprint keyword spotting KWS low CPU memory latency requirement maxpooling loss training guided initializing crossentropy loss t...
[0.010412102565169334, -0.015440134331583977, 0.043549053370952606, 0.07867394387722015, 0.02645268850028515, -0.0025441693142056465, -0.019589930772781372, 0.012536800466477871, -0.011520572938024998, -0.03392830118536949, -0.07016239315271378, -0.06755037605762482, 0.007011089939624071, 0.05525220185518265, -0.008314...
1,824
1,824
['Hamid Reza Hassanzadeh', 'Ying Sha', 'May D. Wang']
1705.03508v1
Multiple cause-of-death data provides a valuable source of information that can be used to enhance health standards by predicting health related trajectories in societies with large populations. These data are often available in large quantities across U.S. states and require Big Data techniques to uncover complex hidd...
DeepDeath: Learning to Predict the Underlying Cause of Death with Big Data
2,017
http://arxiv.org/pdf/1705.03508v1
Title DeepDeath Learning Predict Underlying Cause Death Big Data Summary Multiple causeofdeath data provides valuable source information used enhance health standard predicting health related trajectory society large population data often available large quantity across US state require Big Data technique uncover compl...
[0.008331306278705597, 0.10236362367868423, -0.023596465587615967, -0.031271081417798996, 0.03516004607081413, 0.024835936725139618, 0.009174756705760956, 0.01896328292787075, -0.026599928736686707, 0.01684981770813465, 0.0715896338224411, 0.031552813947200775, -0.023580633103847504, 0.055329810827970505, -0.0051669990...
1,825
1,825
['Matthias Plappert', 'Christian Mandery', 'Tamim Asfour']
1705.06400v1
Linking human whole-body motion and natural language is of great interest for the generation of semantic representations of observed human behaviors as well as for the generation of robot behaviors based on natural language input. While there has been a large body of research in this area, most approaches that exist to...
Learning a bidirectional mapping between human whole-body motion and natural language using deep recurrent neural networks
2,017
http://arxiv.org/pdf/1705.06400v1
Title Learning bidirectional mapping human wholebody motion natural language using deep recurrent neural network Summary Linking human wholebody motion natural language great interest generation semantic representation observed human behavior well generation robot behavior based natural language input large body resear...
[0.027801254764199257, 0.02275240607559681, -0.0030444092117249966, 0.04590127244591713, -0.00805177353322506, 0.027435606345534325, 0.0034732650965452194, 0.003980742767453194, -0.024588244035840034, -0.03487956151366234, -0.014160163700580597, -0.0657811313867569, 0.031440719962120056, 0.05025310441851616, 0.03421469...
1,826
1,826
['Xuezhe Ma', 'Pengcheng Yin', 'Jingzhou Liu', 'Graham Neubig', 'Eduard Hovy']
1705.07136v3
Reward augmented maximum likelihood (RAML), a simple and effective learning framework to directly optimize towards the reward function in structured prediction tasks, has led to a number of impressive empirical successes. RAML incorporates task-specific reward by performing maximum-likelihood updates on candidate outpu...
Softmax Q-Distribution Estimation for Structured Prediction: A Theoretical Interpretation for RAML
2,017
http://arxiv.org/pdf/1705.07136v3
Title Softmax QDistribution Estimation Structured Prediction Theoretical Interpretation RAML Summary Reward augmented maximum likelihood RAML simple effective learning framework directly optimize towards reward function structured prediction task led number impressive empirical success RAML incorporates taskspecific re...
[0.005534765310585499, -0.014428973197937012, -0.008822360076010227, -0.0013082973891869187, 0.03529583290219307, 0.003050207858905196, -0.02613646164536476, 0.025901872664690018, -0.061393942683935165, -0.03973569720983505, -0.041300397366285324, 0.015440328046679497, -0.0006862009176984429, 0.02623654156923294, -0.00...
1,827
1,827
['Vlad Niculae', 'Mathieu Blondel']
1705.07704v2
Modern neural networks are often augmented with an attention mechanism, which tells the network where to focus within the input. We propose in this paper a new framework for sparse and structured attention, building upon a smoothed max operator. We show that the gradient of this operator defines a mapping from real val...
A Regularized Framework for Sparse and Structured Neural Attention
2,017
http://arxiv.org/pdf/1705.07704v2
Title Regularized Framework Sparse Structured Neural Attention Summary Modern neural network often augmented attention mechanism tell network focus within input propose paper new framework sparse structured attention building upon smoothed max operator show gradient operator defines mapping real value probability suita...
[0.04281168058514595, -0.008051825687289238, 0.007989066652953625, 0.043577779084444046, 0.00376615347340703, -0.009315427392721176, -0.015790650621056557, -0.012781123630702496, -0.037789326161146164, -0.03239327296614647, 0.027258742600679398, 0.014095931313931942, 0.020341506227850914, 0.033959705382585526, 0.048244...
1,828
1,828
['Graham Neubig', 'Yoav Goldberg', 'Chris Dyer']
1705.07860v1
Dynamic neural network toolkits such as PyTorch, DyNet, and Chainer offer more flexibility for implementing models that cope with data of varying dimensions and structure, relative to toolkits that operate on statically declared computations (e.g., TensorFlow, CNTK, and Theano). However, existing toolkits - both static...
On-the-fly Operation Batching in Dynamic Computation Graphs
2,017
http://arxiv.org/pdf/1705.07860v1
Title Onthefly Operation Batching Dynamic Computation Graphs Summary Dynamic neural network toolkits PyTorch DyNet Chainer offer flexibility implementing model cope data varying dimension structure relative toolkits operate statically declared computation eg TensorFlow CNTK Theano However existing toolkits static dynam...
[-0.020121920853853226, -0.02169102057814598, -0.027820486575365067, -0.01208677887916565, -0.04710698500275612, -0.011611077934503555, 0.050652824342250824, -0.02984658256173134, -0.06095295771956444, -0.010977407917380333, -0.004650963470339775, 0.013388720341026783, 0.035430897027254105, 0.07192371785640717, 0.04438...
1,829
1,829
['Andros Tjandra', 'Sakriani Sakti', 'Ruli Manurung', 'Mirna Adriani', 'Satoshi Nakamura']
1706.02222v1
Recurrent Neural Networks (RNNs), which are a powerful scheme for modeling temporal and sequential data need to capture long-term dependencies on datasets and represent them in hidden layers with a powerful model to capture more information from inputs. For modeling long-term dependencies in a dataset, the gating mecha...
Gated Recurrent Neural Tensor Network
2,017
http://arxiv.org/pdf/1706.02222v1
Title Gated Recurrent Neural Tensor Network Summary Recurrent Neural Networks RNNs powerful scheme modeling temporal sequential data need capture longterm dependency datasets represent hidden layer powerful model capture information input modeling longterm dependency dataset gating mechanism concept help RNNs remember ...
[0.017721425741910934, 0.046132899820804596, -0.0054885572753846645, 0.044502150267362595, -0.039388127624988556, 0.011313661932945251, 0.051453735679388046, -0.009717851877212524, -0.014232832007110119, -0.033683110028505325, -0.02335463836789131, -0.0509352870285511, 0.04812736436724663, 0.05160074308514595, 0.005500...
1,830
1,830
['Franziska Horn']
1706.02496v1
With a simple architecture and the ability to learn meaningful word embeddings efficiently from texts containing billions of words, word2vec remains one of the most popular neural language models used today. However, as only a single embedding is learned for every word in the vocabulary, the model fails to optimally re...
Context encoders as a simple but powerful extension of word2vec
2,017
http://arxiv.org/pdf/1706.02496v1
Title Context encoders simple powerful extension word2vec Summary simple architecture ability learn meaningful word embeddings efficiently text containing billion word word2vec remains one popular neural language model used today However single embedding learned every word vocabulary model fails optimally represent wor...
[0.03237437084317207, 0.036730531603097916, 0.026881271973252296, 0.0899711400270462, -0.002664008643478155, 0.04068823531270027, -0.009313885122537613, 0.018034759908914566, -0.023744024336338043, -0.03521965071558952, -0.006245879922062159, -0.016823990270495415, 0.005164702422916889, 0.03268356993794441, 0.037073027...
1,831
1,831
['Yizhe Zhang', 'Zhe Gan', 'Kai Fan', 'Zhi Chen', 'Ricardo Henao', 'Dinghan Shen', 'Lawrence Carin']
1706.03850v3
The Generative Adversarial Network (GAN) has achieved great success in generating realistic (real-valued) synthetic data. However, convergence issues and difficulties dealing with discrete data hinder the applicability of GAN to text. We propose a framework for generating realistic text via adversarial training. We emp...
Adversarial Feature Matching for Text Generation
2,017
http://arxiv.org/pdf/1706.03850v3
Title Adversarial Feature Matching Text Generation Summary Generative Adversarial Network GAN achieved great success generating realistic realvalued synthetic data However convergence issue difficulty dealing discrete data hinder applicability GAN text propose framework generating realistic text via adversarial trainin...
[0.04940181225538254, 0.1137358620762825, -0.005371314473450184, 0.04041068255901337, -0.021193934604525566, -0.015873145312070847, 0.00024993170518428087, 0.006878057029098272, -0.028045745566487312, -0.006333008408546448, -0.0033882923889905214, -0.02059255540370941, -0.001305832527577877, 0.06811115890741348, 0.0394...
1,832
1,832
['Kelsey MacMillan', 'James D. Wilson']
1706.05084v2
Topic models have been extensively used to organize and interpret the contents of large, unstructured corpora of text documents. Although topic models often perform well on traditional training vs. test set evaluations, it is often the case that the results of a topic model do not align with human interpretation. This ...
Topic supervised non-negative matrix factorization
2,017
http://arxiv.org/pdf/1706.05084v2
Title Topic supervised nonnegative matrix factorization Summary Topic model extensively used organize interpret content large unstructured corpus text document Although topic model often perform well traditional training v test set evaluation often case result topic model align human interpretation interpretability fal...
[-0.0029168743640184402, -0.013864993117749691, -0.002284037182107568, 0.011861063539981842, -0.025146737694740295, 0.019236277788877487, 0.005330471787601709, 0.012753811664879322, -0.04505971446633339, -0.03979729488492012, -0.03409549221396446, 0.019031653180718422, -0.013148358091711998, 0.04875137284398079, -0.013...
1,833
1,833
['Chung-Cheng Chiu', 'Dieterich Lawson', 'Yuping Luo', 'George Tucker', 'Kevin Swersky', 'Ilya Sutskever', 'Navdeep Jaitly']
1706.06428v1
Generative models have long been the dominant approach for speech recognition. The success of these models however relies on the use of sophisticated recipes and complicated machinery that is not easily accessible to non-practitioners. Recent innovations in Deep Learning have given rise to an alternative - discriminati...
An online sequence-to-sequence model for noisy speech recognition
2,017
http://arxiv.org/pdf/1706.06428v1
Title online sequencetosequence model noisy speech recognition Summary Generative model long dominant approach speech recognition success model however relies use sophisticated recipe complicated machinery easily accessible nonpractitioners Recent innovation Deep Learning given rise alternative discriminative model cal...
[0.0326002836227417, 0.11502310633659363, 0.0036454300861805677, -0.0018506776541471481, -0.02494509518146515, -0.012069573625922203, 0.04079924523830414, -0.010102986358106136, -0.01342661865055561, -0.023803899064660072, 0.00980338640511036, -0.03204767405986786, 0.03243255242705345, 0.07158708572387695, -0.028992261...
1,834
1,834
['Karl Moritz Hermann', 'Felix Hill', 'Simon Green', 'Fumin Wang', 'Ryan Faulkner', 'Hubert Soyer', 'David Szepesvari', 'Wojciech Marian Czarnecki', 'Max Jaderberg', 'Denis Teplyashin', 'Marcus Wainwright', 'Chris Apps', 'Demis Hassabis', 'Phil Blunsom']
1706.06551v2
We are increasingly surrounded by artificially intelligent technology that takes decisions and executes actions on our behalf. This creates a pressing need for general means to communicate with, instruct and guide artificial agents, with human language the most compelling means for such communication. To achieve this i...
Grounded Language Learning in a Simulated 3D World
2,017
http://arxiv.org/pdf/1706.06551v2
Title Grounded Language Learning Simulated 3D World Summary increasingly surrounded artificially intelligent technology take decision executes action behalf creates pressing need general mean communicate instruct guide artificial agent human language compelling mean communication achieve scalable fashion agent must abl...
[0.06550467014312744, 0.0003195333993062377, -0.026278335601091385, 0.02365860715508461, -0.019051386043429375, 0.005904109217226505, -0.0008557568653486669, -0.011147919110953808, -0.028864098712801933, -0.04711990803480148, -0.017528412863612175, 0.028209909796714783, 0.02906840853393078, 0.11125624924898148, 0.02857...
1,835
1,835
['Wenbo Hu', 'Lifeng Hua', 'Lei Li', 'Hang Su', 'Tian Wang', 'Ning Chen', 'Bo Zhang']
1707.00117v3
This paper presents a Semantic Attribute Modulation (SAM) for language modeling and style variation. The semantic attribute modulation includes various document attributes, such as titles, authors, and document categories. We consider two types of attributes, (title attributes and category attributes), and a flexible a...
SAM: Semantic Attribute Modulation for Language Modeling and Style Variation
2,017
http://arxiv.org/pdf/1707.00117v3
Title SAM Semantic Attribute Modulation Language Modeling Style Variation Summary paper present Semantic Attribute Modulation SAM language modeling style variation semantic attribute modulation includes various document attribute title author document category consider two type attribute title attribute category attrib...
[0.03976268693804741, 0.003608367405831814, -0.017103103920817375, 0.028654415160417557, 0.015316984616219997, -0.021936994045972824, 0.03512682020664215, -0.006940001156181097, -0.09757973998785019, -0.03323424234986305, -0.04097335413098335, 0.04158739000558853, 0.0359344556927681, 0.026433439925312996, 0.00851754844...
1,836
1,836
['Junxian He', 'Zhiting Hu', 'Taylor Berg-Kirkpatrick', 'Ying Huang', 'Eric P. Xing']
1707.00206v1
Correlated topic modeling has been limited to small model and problem sizes due to their high computational cost and poor scaling. In this paper, we propose a new model which learns compact topic embeddings and captures topic correlations through the closeness between the topic vectors. Our method enables efficient inf...
Efficient Correlated Topic Modeling with Topic Embedding
2,017
http://arxiv.org/pdf/1707.00206v1
Title Efficient Correlated Topic Modeling Topic Embedding Summary Correlated topic modeling limited small model problem size due high computational cost poor scaling paper propose new model learns compact topic embeddings capture topic correlation closeness topic vector method enables efficient inference lowdimensional...
[0.01663253828883171, 0.030743500217795372, -0.010040831752121449, -0.003768146736547351, -0.012060718610882759, 0.02037402242422104, 0.010165836662054062, 0.015913911163806915, -0.008548949845135212, -0.037382449954748154, -0.022426512092351913, 0.04258076101541519, -0.011049528606235981, 0.007023899350315332, -0.0358...
1,837
1,837
['Bin Wang', 'Zhijian Ou']
1707.07240v3
Trans-dimensional random field language models (TRF LMs) have recently been introduced, where sentences are modeled as a collection of random fields. The TRF approach has been shown to have the advantages of being computationally more efficient in inference than LSTM LMs with close performance and being able to flexibl...
Language modeling with Neural trans-dimensional random fields
2,017
http://arxiv.org/pdf/1707.07240v3
Title Language modeling Neural transdimensional random field Summary Transdimensional random field language model TRF LMs recently introduced sentence modeled collection random field TRF approach shown advantage computationally efficient inference LSTM LMs close performance able flexibly integrating rich feature paper ...
[0.04054650664329529, 0.027760276570916176, 0.02207040973007679, 0.0742020308971405, -0.01302926242351532, -0.016485195606946945, -0.005472550168633461, 0.013007004745304585, -0.02622521109879017, -0.07754658162593842, -0.011160326190292835, -0.0649813860654831, 0.003937631845474243, 0.03651946038007736, 0.023750176653...
1,838
1,838
['Carolin Lawrence', 'Artem Sokolov', 'Stefan Riezler']
1707.09118v3
The goal of counterfactual learning for statistical machine translation (SMT) is to optimize a target SMT system from logged data that consist of user feedback to translations that were predicted by another, historic SMT system. A challenge arises by the fact that risk-averse commercial SMT systems deterministically lo...
Counterfactual Learning from Bandit Feedback under Deterministic Logging: A Case Study in Statistical Machine Translation
2,017
http://arxiv.org/pdf/1707.09118v3
Title Counterfactual Learning Bandit Feedback Deterministic Logging Case Study Statistical Machine Translation Summary goal counterfactual learning statistical machine translation SMT optimize target SMT system logged data consist user feedback translation predicted another historic SMT system challenge arises fact ris...
[0.054895851761102676, 0.03990679606795311, 0.002057908568531275, 0.010578499175608158, -0.010700840502977371, -0.003152301302179694, 0.011955179274082184, 0.03063059039413929, -0.03695640340447426, -0.03831270709633827, 0.014317121356725693, -0.002317818347364664, -0.009787265211343765, 0.08596131205558777, 0.04238010...
1,839
1,839
['Ekaterina Lobacheva', 'Nadezhda Chirkova', 'Dmitry Vetrov']
1708.00077v1
Recurrent neural networks show state-of-the-art results in many text analysis tasks but often require a lot of memory to store their weights. Recently proposed Sparse Variational Dropout eliminates the majority of the weights in a feed-forward neural network without significant loss of quality. We apply this technique ...
Bayesian Sparsification of Recurrent Neural Networks
2,017
http://arxiv.org/pdf/1708.00077v1
Title Bayesian Sparsification Recurrent Neural Networks Summary Recurrent neural network show stateoftheart result many text analysis task often require lot memory store weight Recently proposed Sparse Variational Dropout eliminates majority weight feedforward neural network without significant loss quality apply techn...
[0.02522731013596058, 0.09828785061836243, -0.004333547316491604, 0.037238575518131256, -0.019507475197315216, -0.007147683762013912, -0.009073835797607899, 0.03292909637093544, -0.034033484756946564, -0.030231911689043045, 0.005273398943245411, -0.01396992802619934, 0.04306242987513542, 0.05029163509607315, -0.0124107...
1,840
1,840
['Benjamin J. Lengerich', 'Andrew L. Maas', 'Christopher Potts']
1708.00112v2
Knowledge graphs are a versatile framework to encode richly structured data relationships, but it not always apparent how to combine these with existing entity representations. Methods for retrofitting pre-trained entity representations to the structure of a knowledge graph typically assume that entities are embedded i...
Retrofitting Distributional Embeddings to Knowledge Graphs with Functional Relations
2,017
http://arxiv.org/pdf/1708.00112v2
Title Retrofitting Distributional Embeddings Knowledge Graphs Functional Relations Summary Knowledge graph versatile framework encode richly structured data relationship always apparent combine existing entity representation Methods retrofitting pretrained entity representation structure knowledge graph typically assum...
[0.03275388106703758, 0.05236857384443283, 0.00025364645989611745, 0.03261345252394676, 0.02475053444504738, -0.005807383917272091, -0.05302572622895241, 0.027501754462718964, 0.053151872009038925, -0.003465380286797881, 0.011728880926966667, 0.0014357449254021049, -0.012242274358868599, 0.058934278786182404, 0.0208486...
1,841
1,841
['Ramesh Nallapati', 'Igor Melnyk', 'Abhishek Kumar', 'Bowen Zhou']
1708.00308v1
We present a new topic model that generates documents by sampling a topic for one whole sentence at a time, and generating the words in the sentence using an RNN decoder that is conditioned on the topic of the sentence. We argue that this novel formalism will help us not only visualize and model the topical discourse s...
SenGen: Sentence Generating Neural Variational Topic Model
2,017
http://arxiv.org/pdf/1708.00308v1
Title SenGen Sentence Generating Neural Variational Topic Model Summary present new topic model generates document sampling topic one whole sentence time generating word sentence using RNN decoder conditioned topic sentence argue novel formalism help u visualize model topical discourse structure document better also po...
[0.05194584280252457, 0.06273024529218674, -0.013617176562547684, 0.02858302742242813, -0.044935643672943115, 0.0011296090669929981, -0.019889336079359055, -0.035215046256780624, -0.06657394766807556, -0.031558871269226074, 0.03725748509168625, 0.010543186217546463, 0.01314430683851242, 0.07080081105232239, 0.018816549...
1,842
1,842
['Lee Gao', 'Ronghuo Zheng']
1708.03052v1
Embarrassingly (communication-free) parallel Markov chain Monte Carlo (MCMC) methods are commonly used in learning graphical models. However, MCMC cannot be directly applied in learning topic models because of the quasi-ergodicity problem caused by multimodal distribution of topics. In this paper, we develop an embarra...
Communication-Free Parallel Supervised Topic Models
2,017
http://arxiv.org/pdf/1708.03052v1
Title CommunicationFree Parallel Supervised Topic Models Summary Embarrassingly communicationfree parallel Markov chain Monte Carlo MCMC method commonly used learning graphical model However MCMC cannot directly applied learning topic model quasiergodicity problem caused multimodal distribution topic paper develop emba...
[0.0258750282227993, 0.019592249765992165, 0.005679073743522167, 0.004068332724273205, -0.02884518727660179, 0.025453658774495125, 0.030736198648810387, -0.0010708080371841788, -0.07014484703540802, -0.027431946247816086, -0.011804269626736641, 0.03456386551260948, -0.004455603193491697, 0.022660553455352783, -0.024812...
1,843
1,843
['Prathusha Kameswara Sarma', 'Bill Sethares']
1708.03995v1
Word embeddings are representations of individual words of a text document in a vector space and they are often use- ful for performing natural language pro- cessing tasks. Current state of the art al- gorithms for learning word embeddings learn vector representations from large corpora of text documents in an unsu- pe...
Sentiment Analysis by Joint Learning of Word Embeddings and Classifier
2,017
http://arxiv.org/pdf/1708.03995v1
Title Sentiment Analysis Joint Learning Word Embeddings Classifier Summary Word embeddings representation individual word text document vector space often use ful performing natural language pro cessing task Current state art al gorithms learning word embeddings learn vector representation large corpus text document un...
[0.027219584211707115, 0.010280968621373177, 0.010779744945466518, 0.0676899328827858, -0.009344675578176975, 0.011329025961458683, -0.02473275177180767, 0.002063492313027382, 0.03428564593195915, -0.08381877839565277, -0.039447933435440063, 0.0078089372254908085, 0.009680327959358692, 0.017475370317697525, -0.03432043...
1,844
1,844
['Yizhe Zhang', 'Dinghan Shen', 'Guoyin Wang', 'Zhe Gan', 'Ricardo Henao', 'Lawrence Carin']
1708.04729v3
Learning latent representations from long text sequences is an important first step in many natural language processing applications. Recurrent Neural Networks (RNNs) have become a cornerstone for this challenging task. However, the quality of sentences during RNN-based decoding (reconstruction) decreases with the leng...
Deconvolutional Paragraph Representation Learning
2,017
http://arxiv.org/pdf/1708.04729v3
Title Deconvolutional Paragraph Representation Learning Summary Learning latent representation long text sequence important first step many natural language processing application Recurrent Neural Networks RNNs become cornerstone challenging task However quality sentence RNNbased decoding reconstruction decrease length...
[0.041751861572265625, 0.05148877948522568, 0.014124581590294838, 0.0679616779088974, -0.027368532493710518, -0.002680003410205245, -0.00624211085960269, -0.007884124293923378, -0.022914228960871696, -0.04875979200005531, 0.028614891692996025, -0.009732588194310665, 0.04433976486325264, 0.011811420321464539, 0.00033976...
1,845
1,845
['Vikramjit Mitra', 'Horacio Franco']
1708.09516v1
Unseen data conditions can inflict serious performance degradation on systems relying on supervised machine learning algorithms. Because data can often be unseen, and because traditional machine learning algorithms are trained in a supervised manner, unsupervised adaptation techniques must be used to adapt the model to...
Leveraging Deep Neural Network Activation Entropy to cope with Unseen Data in Speech Recognition
2,017
http://arxiv.org/pdf/1708.09516v1
Title Leveraging Deep Neural Network Activation Entropy cope Unseen Data Speech Recognition Summary Unseen data condition inflict serious performance degradation system relying supervised machine learning algorithm data often unseen traditional machine learning algorithm trained supervised manner unsupervised adaptatio...
[-0.01947842352092266, 0.07939974218606949, 0.00484080146998167, 0.041968099772930145, 0.007223476190119982, -0.014769133180379868, 0.04008758068084717, -0.0003933074476663023, -0.06650547683238983, 0.006552563514560461, -0.050352804362773895, 0.007729894481599331, 0.04533961042761803, 0.024134313687682152, 0.016623895...
1,846
1,846
['Dinghan Shen', 'Yizhe Zhang', 'Ricardo Henao', 'Qinliang Su', 'Lawrence Carin']
1709.07109v3
A latent-variable model is introduced for text matching, inferring sentence representations by jointly optimizing generative and discriminative objectives. To alleviate typical optimization challenges in latent-variable models for text, we employ deconvolutional networks as the sequence decoder (generator), providing l...
Deconvolutional Latent-Variable Model for Text Sequence Matching
2,017
http://arxiv.org/pdf/1709.07109v3
Title Deconvolutional LatentVariable Model Text Sequence Matching Summary latentvariable model introduced text matching inferring sentence representation jointly optimizing generative discriminative objective alleviate typical optimization challenge latentvariable model text employ deconvolutional network sequence deco...
[0.043444596230983734, 0.08836998045444489, 0.011503105983138084, 0.07553686946630478, -0.05090409144759178, 0.012517988681793213, -0.015127968974411488, 0.0027745761908590794, -0.014369658194482327, -0.02736745961010456, 0.02441183477640152, -0.0007418266031891108, -0.0061560506001114845, 0.04577536880970001, -0.01053...
1,847
1,847
['Dinghan Shen', 'Martin Renqiang Min', 'Yitong Li', 'Lawrence Carin']
1709.08294v1
Convolutional neural networks (CNNs) have recently emerged as a popular building block for natural language processing (NLP). Despite their success, most existing CNN models employed in NLP are not expressive enough, in the sense that all input sentences share the same learned (and static) set of filters. Motivated by ...
Adaptive Convolutional Filter Generation for Natural Language Understanding
2,017
http://arxiv.org/pdf/1709.08294v1
Title Adaptive Convolutional Filter Generation Natural Language Understanding Summary Convolutional neural network CNNs recently emerged popular building block natural language processing NLP Despite success existing CNN model employed NLP expressive enough sense input sentence share learned static set filter Motivated...
[0.07217704504728317, 0.0017360455822199583, -0.014277510344982147, 0.03864992782473564, -0.0038339614402502775, 0.004525600932538509, 0.018498515710234642, -0.01094012800604105, -0.047462958842515945, -0.07622891664505005, -0.011773571372032166, 0.0771334171295166, -0.04495719447731972, 0.07067780196666718, -0.0262536...
1,848
1,848
['Maja Rudolph', 'Francisco Ruiz', 'Susan Athey', 'David Blei']
1709.10367v1
Word embeddings are a powerful approach for analyzing language, and exponential family embeddings (EFE) extend them to other types of data. Here we develop structured exponential family embeddings (S-EFE), a method for discovering embeddings that vary across related groups of data. We study how the word usage of U.S. C...
Structured Embedding Models for Grouped Data
2,017
http://arxiv.org/pdf/1709.10367v1
Title Structured Embedding Models Grouped Data Summary Word embeddings powerful approach analyzing language exponential family embeddings EFE extend type data develop structured exponential family embeddings SEFE method discovering embeddings vary across related group data study word usage US Congressional speech varie...
[0.001469915034249425, 0.07553967833518982, -0.02832489274442196, 0.05755311995744705, 0.020636675879359245, 0.020734012126922607, 0.020959509536623955, 0.0268073882907629, -0.012439632788300514, -0.05452030524611473, -0.0027075547259300947, -0.02878692001104355, 0.021389391273260117, 0.04713856056332588, 0.00309216580...
1,849
1,849
['Finn Årup Nielsen']
1710.04099v1
I present a web service for querying an embedding of entities in the Wikidata knowledge graph. The embedding is trained on the Wikidata dump using Gensim's Word2Vec implementation and a simple graph walk. A REST API is implemented. Together with the Wikidata API the web service exposes a multilingual resource for over ...
Wembedder: Wikidata entity embedding web service
2,017
http://arxiv.org/pdf/1710.04099v1
Title Wembedder Wikidata entity embedding web service Summary present web service querying embedding entity Wikidata knowledge graph embedding trained Wikidata dump using Gensims Word2Vec implementation simple graph walk REST API implemented Together Wikidata API web service expose multilingual resource 600000 Wikidata...
[0.044290948659181595, 0.062054894864559174, 0.0038098134100437164, 0.04812001809477806, -0.005733951926231384, 0.035716474056243896, -0.018254484981298447, 0.03765834867954254, 0.04221441224217415, -0.02258770540356636, 0.0418480709195137, 0.009084900841116905, 0.021865935996174812, 0.03681518882513046, 0.023787533864...
1,850
1,850
['Patrick O. Perry', 'Kenneth Benoit']
1710.08963v1
Probabilistic methods for classifying text form a rich tradition in machine learning and natural language processing. For many important problems, however, class prediction is uninteresting because the class is known, and instead the focus shifts to estimating latent quantities related to the text, such as affect or id...
Scaling Text with the Class Affinity Model
2,017
http://arxiv.org/pdf/1710.08963v1
Title Scaling Text Class Affinity Model Summary Probabilistic method classifying text form rich tradition machine learning natural language processing many important problem however class prediction uninteresting class known instead focus shift estimating latent quantity related text affect ideology focus one problem i...
[0.019651321694254875, 0.041688378900289536, -0.012474129907786846, 0.006815330125391483, -0.04597922042012215, -0.007052906323224306, 0.02446204423904419, 0.010155964642763138, -0.009768675081431866, -0.04713018611073494, 0.04675336927175522, -0.020734529942274094, 0.015372729860246181, 0.06763906031847, -0.0313959829...
1,851
1,851
['Markus Kliegl', 'Siddharth Goyal', 'Kexin Zhao', 'Kavya Srinet', 'Mohammad Shoeybi']
1710.09026v2
We propose and evaluate new techniques for compressing and speeding up dense matrix multiplications as found in the fully connected and recurrent layers of neural networks for embedded large vocabulary continuous speech recognition (LVCSR). For compression, we introduce and study a trace norm regularization technique f...
Trace norm regularization and faster inference for embedded speech recognition RNNs
2,017
http://arxiv.org/pdf/1710.09026v2
Title Trace norm regularization faster inference embedded speech recognition RNNs Summary propose evaluate new technique compressing speeding dense matrix multiplication found fully connected recurrent layer neural network embedded large vocabulary continuous speech recognition LVCSR compression introduce study trace n...
[-0.012550274841487408, 0.012596708722412586, 0.00814382079988718, 0.0743304193019867, 0.014276435598731041, -0.008364450186491013, 0.002128100488334894, 0.009747951291501522, -0.020834386348724365, -0.012311897240579128, -0.04305514693260193, 0.0013044878141954541, 0.04448592662811279, -0.0024215970188379288, 0.010201...
1,852
1,852
['Long Chen', 'Fajie Yuan', 'Joemon M. Jose', 'Weinan Zhang']
1710.09805v2
Although the word-popularity based negative sampler has shown superb performance in the skip-gram model, the theoretical motivation behind oversampling popular (non-observed) words as negative samples is still not well understood. In this paper, we start from an investigation of the gradient vanishing issue in the skip...
Improving Negative Sampling for Word Representation using Self-embedded Features
2,017
http://arxiv.org/pdf/1710.09805v2
Title Improving Negative Sampling Word Representation using Selfembedded Features Summary Although wordpopularity based negative sampler shown superb performance skipgram model theoretical motivation behind oversampling popular nonobserved word negative sample still well understood paper start investigation gradient va...
[-0.0015123027842491865, -0.028344908729195595, -0.025834977626800537, 0.016172364354133606, -0.025088999420404434, -0.03281921520829201, -0.005570703186094761, 0.001916167326271534, 0.0033314877655357122, -0.0594368539750576, -0.028455771505832672, -0.030560128390789032, 0.01730652153491974, 0.044220149517059326, 0.01...
1,853
1,853
['Andrew K. Lampinen', 'James L. McClelland']
1710.10280v2
Standard deep learning systems require thousands or millions of examples to learn a concept, and cannot integrate new concepts easily. By contrast, humans have an incredible ability to do one-shot or few-shot learning. For instance, from just hearing a word used in a sentence, humans can infer a great deal about it, by...
One-shot and few-shot learning of word embeddings
2,017
http://arxiv.org/pdf/1710.10280v2
Title Oneshot fewshot learning word embeddings Summary Standard deep learning system require thousand million example learn concept cannot integrate new concept easily contrast human incredible ability oneshot fewshot learning instance hearing word used sentence human infer great deal leveraging syntax semantics surrou...
[0.04413899406790733, 0.04056582972407341, 0.023977525532245636, 0.040236201137304306, -0.035199955105781555, -0.023909643292427063, 0.0133059062063694, 0.014272532425820827, 0.0029199945274740458, -0.04140124469995499, 0.03165460377931595, 0.01633337140083313, -0.0034260593820363283, 0.06688492745161057, 0.05582799389...
1,854
1,854
['Li Wan', 'Quan Wang', 'Alan Papir', 'Ignacio Lopez Moreno']
1710.10467v2
In this paper, we propose a new loss function called generalized end-to-end (GE2E) loss, which makes the training of speaker verification models more efficient than our previous tuple-based end-to-end (TE2E) loss function. Unlike TE2E, the GE2E loss function updates the network in a way that emphasizes examples that ar...
Generalized End-to-End Loss for Speaker Verification
2,017
http://arxiv.org/pdf/1710.10467v2
Title Generalized EndtoEnd Loss Speaker Verification Summary paper propose new loss function called generalized endtoend GE2E loss make training speaker verification model efficient previous tuplebased endtoend TE2E loss function Unlike TE2E GE2E loss function update network way emphasizes example difficult verify step...
[0.006510249804705381, 0.027715448290109634, 0.033141691237688065, 0.04983142390847206, -0.016884053125977516, 0.006236733868718147, -0.002364868763834238, -0.0070679825730621815, -0.02444346807897091, -0.011054336093366146, -0.02715938724577427, -0.05207155644893646, 0.0488177053630352, -0.01137388963252306, 0.0405002...
1,855
1,855
['Taku Kato', 'Takahiro Shinozaki']
1711.03689v1
Speech recognition systems have achieved high recognition performance for several tasks. However, the performance of such systems is dependent on the tremendously costly development work of preparing vast amounts of task-matched transcribed speech data for supervised training. The key problem here is the cost of transc...
Reinforcement Learning of Speech Recognition System Based on Policy Gradient and Hypothesis Selection
2,017
http://arxiv.org/pdf/1711.03689v1
Title Reinforcement Learning Speech Recognition System Based Policy Gradient Hypothesis Selection Summary Speech recognition system achieved high recognition performance several task However performance system dependent tremendously costly development work preparing vast amount taskmatched transcribed speech data super...
[0.03870672360062599, 0.005879160016775131, 0.012348457239568233, 0.014990766532719135, -0.01500969473272562, -0.00917383749037981, 0.03786231577396393, 0.01467077899724245, -0.01748782768845558, -0.025669656693935394, -0.05377897992730141, 0.010621750727295876, 0.024248860776424408, 0.0029823067598044872, -0.050133060...
1,856
1,856
['Geng Ji', 'Robert Bamler', 'Erik B. Sudderth', 'Stephan Mandt']
1711.03946v2
Word2vec (Mikolov et al., 2013) has proven to be successful in natural language processing by capturing the semantic relationships between different words. Built on top of single-word embeddings, paragraph vectors (Le and Mikolov, 2014) find fixed-length representations for pieces of text with arbitrary lengths, such a...
Bayesian Paragraph Vectors
2,017
http://arxiv.org/pdf/1711.03946v2
Title Bayesian Paragraph Vectors Summary Word2vec Mikolov et al 2013 proven successful natural language processing capturing semantic relationship different word Built top singleword embeddings paragraph vector Le Mikolov 2014 find fixedlength representation piece text arbitrary length document paragraph sentence work ...
[0.05483843758702278, 0.051504094153642654, 0.012798028998076916, 0.044844143092632294, -0.07368046045303345, 0.0051512024365365505, 0.0028310830239206553, 0.014285609126091003, -0.04275105893611908, -0.07155312597751617, 0.02685491368174553, 0.01302750501781702, 0.03402569144964218, -0.007554207928478718, 0.0039467802...
1,857
1,857
['Chundi Liu', 'Shunan Zhao', 'Maksims Volkovs']
1711.04168v3
We propose a new model for unsupervised document embedding. Leading existing approaches either require complex inference or use recurrent neural networks (RNN) that are difficult to parallelize. We take a different route and develop a convolutional neural network (CNN) embedding model. Our CNN architecture is fully par...
Unsupervised Document Embedding With CNNs
2,017
http://arxiv.org/pdf/1711.04168v3
Title Unsupervised Document Embedding CNNs Summary propose new model unsupervised document embedding Leading existing approach either require complex inference use recurrent neural network RNN difficult parallelize take different route develop convolutional neural network CNN embedding model CNN architecture fully para...
[0.002093688352033496, 0.041458889842033386, 0.029459131881594658, 0.06816870719194412, -0.01173541322350502, 0.013572599738836288, 0.00982437189668417, 0.003039107657968998, 0.01127932220697403, -0.03339962661266327, -0.008581485599279404, 0.07614598423242569, -0.024118779227137566, 0.004130133427679539, -0.0085583515...
1,858
1,858
['Hamideh Hajiabadi', 'Diego Molla-Aliod', 'Reza Monsefi']
1711.05170v1
Ensemble techniques are powerful approaches that combine several weak learners to build a stronger one. As a meta learning framework, ensemble techniques can easily be applied to many machine learning techniques. In this paper we propose a neural network extended with an ensemble loss function for text classification. ...
On Extending Neural Networks with Loss Ensembles for Text Classification
2,017
http://arxiv.org/pdf/1711.05170v1
Title Extending Neural Networks Loss Ensembles Text Classification Summary Ensemble technique powerful approach combine several weak learner build stronger one meta learning framework ensemble technique easily applied many machine learning technique paper propose neural network extended ensemble loss function text clas...
[0.03224543482065201, 0.01455842424184084, -0.0008307452080771327, 0.013932323083281517, -0.0025849719531834126, 0.014135141856968403, 0.02803918346762657, 0.017099430784583092, -0.020022209733724594, -0.044287897646427155, -0.027068141847848892, 0.020289938896894455, 0.02942313626408577, -0.012270474806427956, 0.00275...
1,859
1,859
['Shankar Kumar', 'Michael Nirschl', 'Daniel Holtmann-Rice', 'Hank Liao', 'Ananda Theertha Suresh', 'Felix Yu']
1711.05448v1
Recurrent neural network (RNN) language models (LMs) and Long Short Term Memory (LSTM) LMs, a variant of RNN LMs, have been shown to outperform traditional N-gram LMs on speech recognition tasks. However, these models are computationally more expensive than N-gram LMs for decoding, and thus, challenging to integrate in...
Lattice Rescoring Strategies for Long Short Term Memory Language Models in Speech Recognition
2,017
http://arxiv.org/pdf/1711.05448v1
Title Lattice Rescoring Strategies Long Short Term Memory Language Models Speech Recognition Summary Recurrent neural network RNN language model LMs Long Short Term Memory LSTM LMs variant RNN LMs shown outperform traditional Ngram LMs speech recognition task However model computationally expensive Ngram LMs decoding t...
[-0.018387332558631897, 0.020932385697960854, 0.010654221288859844, 0.07105951756238937, -0.014253768138587475, 0.009234192781150341, -0.018995383754372597, 0.027203528210520744, -0.0438837856054306, -0.0362146757543087, -0.05256757140159607, -0.03639081493020058, 0.05985630303621292, -0.06560838967561722, -0.017292918...
1,860
1,860
['Carolin Lawrence', 'Pratik Gajane', 'Stefan Riezler']
1711.08621v3
Counterfactual learning is a natural scenario to improve web-based machine translation services by offline learning from feedback logged during user interactions. In order to avoid the risk of showing inferior translations to users, in such scenarios mostly exploration-free deterministic logging policies are in place. ...
Counterfactual Learning for Machine Translation: Degeneracies and Solutions
2,017
http://arxiv.org/pdf/1711.08621v3
Title Counterfactual Learning Machine Translation Degeneracies Solutions Summary Counterfactual learning natural scenario improve webbased machine translation service offline learning feedback logged user interaction order avoid risk showing inferior translation user scenario mostly explorationfree deterministic loggin...
[0.057934872806072235, 0.04945463687181473, -0.03653645142912865, 0.015016845427453518, -0.03669194504618645, 0.035493649542331696, 0.028937773779034615, 0.0386984646320343, -0.0169239304959774, -0.03429905325174332, 0.022772174328565598, -0.023378117009997368, 0.02748720534145832, 0.03867005556821823, 0.02654269710183...
1,861
1,861
['Namkyu Jung', 'Hyeong In Choi']
1711.08870v1
This paper proposes the continuous semantic topic embedding model (CSTEM) which finds latent topic variables in documents using continuous semantic distance function between the topics and the words by means of the variational autoencoder(VAE). The semantic distance could be represented by any symmetric bell-shaped geo...
Continuous Semantic Topic Embedding Model Using Variational Autoencoder
2,017
http://arxiv.org/pdf/1711.08870v1
Title Continuous Semantic Topic Embedding Model Using Variational Autoencoder Summary paper proposes continuous semantic topic embedding model CSTEM find latent topic variable document using continuous semantic distance function topic word mean variational autoencoderVAE semantic distance could represented symmetric be...
[0.005712768994271755, 0.03648697957396507, -0.013836881145834923, 0.05831316113471985, -0.03285568952560425, 0.012888927012681961, -0.01041842345148325, -0.025007911026477814, -0.06467417627573013, -0.041342753916978836, 0.041079673916101456, 0.0379306823015213, -0.027151713147759438, 0.08260930329561234, 0.0154790747...
1,862
1,862
['Ziang Xie']
1711.09534v1
Deep learning methods have recently achieved great empirical success on machine translation, dialogue response generation, summarization, and other text generation tasks. At a high level, the technique has been to train end-to-end neural network models consisting of an encoder model to produce a hidden representation o...
Neural Text Generation: A Practical Guide
2,017
http://arxiv.org/pdf/1711.09534v1
Title Neural Text Generation Practical Guide Summary Deep learning method recently achieved great empirical success machine translation dialogue response generation summarization text generation task high level technique train endtoend neural network model consisting encoder model produce hidden representation source t...
[0.05683393031358719, 0.04359658434987068, 0.007223980035632849, 0.05226228013634682, 0.000722604978363961, -0.011020122095942497, 0.006082830019295216, 0.0009537424193695188, -0.004145828075706959, -0.03284343704581261, 0.030935754999518394, -0.012982445769011974, 0.01710236631333828, 0.07365313172340393, 0.0147438943...
1,863
1,863
['Danijar Hafner', 'Alexander Immer', 'Willi Raschkowski', 'Fabian Windheuser']
1711.10327v1
Learning distributed representations of documents has pushed the state-of-the-art in several natural language processing tasks and was successfully applied to the field of recommender systems recently. In this paper, we propose a novel content-based recommender system based on learned representations and a generative m...
Generative Interest Estimation for Document Recommendations
2,017
http://arxiv.org/pdf/1711.10327v1
Title Generative Interest Estimation Document Recommendations Summary Learning distributed representation document pushed stateoftheart several natural language processing task successfully applied field recommender system recently paper propose novel contentbased recommender system based learned representation generat...
[0.04758233204483986, 0.010598143562674522, 0.005483996123075485, -0.0003957902954425663, 0.010069933719933033, -0.007594099268317223, 0.01080897357314825, 0.000659504032228142, 0.014764674007892609, -0.052642859518527985, -0.04858548566699028, 0.010161180049180984, -0.013706736266613007, 0.043165385723114014, -0.01423...
1,864
1,864
['Felix Hieber', 'Tobias Domhan', 'Michael Denkowski', 'David Vilar', 'Artem Sokolov', 'Ann Clifton', 'Matt Post']
1712.05690v1
We describe Sockeye (version 1.12), an open-source sequence-to-sequence toolkit for Neural Machine Translation (NMT). Sockeye is a production-ready framework for training and applying models as well as an experimental platform for researchers. Written in Python and built on MXNet, the toolkit offers scalable training a...
Sockeye: A Toolkit for Neural Machine Translation
2,017
http://arxiv.org/pdf/1712.05690v1
Title Sockeye Toolkit Neural Machine Translation Summary describe Sockeye version 112 opensource sequencetosequence toolkit Neural Machine Translation NMT Sockeye productionready framework training applying model well experimental platform researcher Written Python built MXNet toolkit offer scalable training inference ...
[0.0280758123844862, 0.05107465758919716, -0.0005777163896709681, 0.02478805184364319, -0.03170614317059517, 0.010706084780395031, 0.036393940448760986, -0.010009834542870522, -0.009142615832388401, -0.02097316086292267, -0.014678329229354858, -0.019802788272500038, 0.02958761714398861, 0.06207440048456192, 0.039467848...
1,865
1,865
['Martin Schrimpf', 'Stephen Merity', 'James Bradbury', 'Richard Socher']
1712.07316v1
The process of designing neural architectures requires expert knowledge and extensive trial and error. While automated architecture search may simplify these requirements, the recurrent neural network (RNN) architectures generated by existing methods are limited in both flexibility and components. We propose a domain-s...
A Flexible Approach to Automated RNN Architecture Generation
2,017
http://arxiv.org/pdf/1712.07316v1
Title Flexible Approach Automated RNN Architecture Generation Summary process designing neural architecture requires expert knowledge extensive trial error automated architecture search may simplify requirement recurrent neural network RNN architecture generated existing method limited flexibility component propose dom...
[0.015559309162199497, 0.03889097273349762, -0.027060607448220253, 0.034304603934288025, -0.02538754977285862, -0.04383179917931557, 0.035753991454839706, -0.017985699698328972, -0.03760603815317154, -0.01329888217151165, 0.0033841088879853487, -0.016549497842788696, 0.016938967630267143, 0.07769522070884705, 0.0053152...
1,866
1,866
['Cedric De Boom', 'Thomas Demeester', 'Bart Dhoedt']
1801.00632v2
Recurrent neural networks are nowadays successfully used in an abundance of applications, going from text, speech and image processing to recommender systems. Backpropagation through time is the algorithm that is commonly used to train these networks on specific tasks. Many deep learning frameworks have their own imple...
Character-level Recurrent Neural Networks in Practice: Comparing Training and Sampling Schemes
2,018
http://arxiv.org/pdf/1801.00632v2
Title Characterlevel Recurrent Neural Networks Practice Comparing Training Sampling Schemes Summary Recurrent neural network nowadays successfully used abundance application going text speech image processing recommender system Backpropagation time algorithm commonly used train network specific task Many deep learning ...
[0.04639318585395813, 0.033681534230709076, -0.006376951467245817, 0.022360164672136307, -0.023306183516979218, -0.03996116295456886, 0.0214530099183321, 0.0249727051705122, -0.03346319496631622, -0.03956788405776024, -0.004901972599327564, -0.0584704726934433, 0.044203270226716995, 0.054446566849946976, 0.024794671684...
1,867
1,867
['Sahil Garg', 'Greg Ver Steeg', 'Aram Galstyan']
1801.03911v2
Natural language processing often involves computations with semantic or syntactic graphs to facilitate sophisticated reasoning based on structural relationships. While convolution kernels provide a powerful tool for comparing graph structure based on node (word) level relationships, they are difficult to customize and...
Stochastic Learning of Nonstationary Kernels for Natural Language Modeling
2,018
http://arxiv.org/pdf/1801.03911v2
Title Stochastic Learning Nonstationary Kernels Natural Language Modeling Summary Natural language processing often involves computation semantic syntactic graph facilitate sophisticated reasoning based structural relationship convolution kernel provide powerful tool comparing graph structure based node word level rela...
[0.05502008646726608, 0.004777956288307905, -0.00746744591742754, 0.046942200511693954, -0.06276801973581314, -0.006146593950688839, -0.007640775293111801, 0.02029438130557537, 0.046715181320905685, -0.07208866626024246, 0.02152264304459095, 0.00586142810061574, 0.003959172870963812, 0.07116866111755371, 0.005181891378...
1,868
1,868
['Quan Hoang']
1801.04813v1
This project explores several Machine Learning methods to predict movie genres based on plot summaries. Naive Bayes, Word2Vec+XGBoost and Recurrent Neural Networks are used for text classification, while K-binary transformation, rank method and probabilistic classification with learned probability threshold are employe...
Predicting Movie Genres Based on Plot Summaries
2,018
http://arxiv.org/pdf/1801.04813v1
Title Predicting Movie Genres Based Plot Summaries Summary project explores several Machine Learning method predict movie genre based plot summary Naive Bayes Word2VecXGBoost Recurrent Neural Networks used text classification Kbinary transformation rank method probabilistic classification learned probability threshold ...
[0.06283286213874817, 0.028008945286273956, -0.017387066036462784, 0.013933061622083187, -0.002621521707624197, 0.0026867196429520845, 0.01868574693799019, 0.03161279857158661, -0.030902236700057983, -0.033350467681884766, -0.007574534974992275, -0.01510925404727459, 0.013027451932430267, 0.08963310718536377, -0.021866...
1,869
1,869
['Robert Giaquinto', 'Arindam Banerjee']
1801.04958v1
Topic modeling enables exploration and compact representation of a corpus. The CaringBridge (CB) dataset is a massive collection of journals written by patients and caregivers during a health crisis. Topic modeling on the CB dataset, however, is challenging due to the asynchronous nature of multiple authors writing abo...
Topic Modeling on Health Journals with Regularized Variational Inference
2,018
http://arxiv.org/pdf/1801.04958v1
Title Topic Modeling Health Journals Regularized Variational Inference Summary Topic modeling enables exploration compact representation corpus CaringBridge CB dataset massive collection journal written patient caregiver health crisis Topic modeling CB dataset however challenging due asynchronous nature multiple author...
[0.04773436114192009, 0.09448570013046265, -0.01778283156454563, -0.031360525637865067, -0.008221045136451721, 0.007884583435952663, 0.039091989398002625, -0.016894899308681488, -0.03013204224407673, 0.006345320958644152, 0.028301481157541275, 0.023323187604546547, 0.008876003324985504, 0.06968356668949127, -0.01607305...
1,870
1,870
['W. James Murdoch', 'Peter J. Liu', 'Bin Yu']
1801.05453v1
The driving force behind the recent success of LSTMs has been their ability to learn complex and non-linear relationships. Consequently, our inability to describe these relationships has led to LSTMs being characterized as black boxes. To this end, we introduce contextual decomposition (CD), an interpretation algorithm...
Beyond Word Importance: Contextual Decomposition to Extract Interactions from LSTMs
2,018
http://arxiv.org/pdf/1801.05453v1
Title Beyond Word Importance Contextual Decomposition Extract Interactions LSTMs Summary driving force behind recent success LSTMs ability learn complex nonlinear relationship Consequently inability describe relationship led LSTMs characterized black box end introduce contextual decomposition CD interpretation algorith...
[0.028604183346033096, 0.016535114496946335, -0.014301621355116367, 0.04364161193370819, -0.049952831119298935, 0.013853798620402813, -0.0002316530590178445, 0.0070762657560408115, -0.010784808546304703, -0.06728674471378326, -0.03651157394051552, 0.0008763070800341666, 0.00823767390102148, 0.056905243545770645, -0.023...
1,871
1,871
['Jeremy Howard', 'Sebastian Ruder']
1801.06146v1
Transfer learning has revolutionized computer vision, but existing approaches in NLP still require task-specific modifications and training from scratch. We propose Fine-tuned Language Models (FitLaM), an effective transfer learning method that can be applied to any task in NLP, and introduce techniques that are key fo...
Fine-tuned Language Models for Text Classification
2,018
http://arxiv.org/pdf/1801.06146v1
Title Finetuned Language Models Text Classification Summary Transfer learning revolutionized computer vision existing approach NLP still require taskspecific modification training scratch propose Finetuned Language Models FitLaM effective transfer learning method applied task NLP introduce technique key finetuning stat...
[0.04873574152588844, 0.02025151066482067, -0.038771022111177444, 0.05725076049566269, -0.021365297958254814, 0.03602663800120354, 0.00822831317782402, 0.015587398782372475, -0.004068480338901281, -0.09304343163967133, -0.018755201250314713, 0.02595602348446846, 0.025951774790883064, 0.03368603065609932, -0.01299636624...
1,872
1,872
['Linyuan Gong', 'Ruyi Ji']
1801.06287v1
TextCNN, the convolutional neural network for text, is a useful deep learning algorithm for sentence classification tasks such as sentiment analysis and question classification. However, neural networks have long been known as black boxes because interpreting them is a challenging task. Researchers have developed sever...
What Does a TextCNN Learn?
2,018
http://arxiv.org/pdf/1801.06287v1
Title TextCNN Learn Summary TextCNN convolutional neural network text useful deep learning algorithm sentence classification task sentiment analysis question classification However neural network long known black box interpreting challenging task Researchers developed several tool understand CNN image classification de...
[0.048387061804533005, 0.04427165165543556, 0.008050741627812386, 0.0726664662361145, -0.04819415882229805, 0.022649478167295456, 0.02912021428346634, -0.0037591082509607077, -0.01508322823792696, -0.03471262380480766, 0.008230684325098991, 0.03237614780664444, 0.020188363268971443, 0.06116760894656181, 0.0057859327644...
1,873
1,873
['Michael Bloodgood']
1801.07875v1
This paper investigates and evaluates support vector machine active learning algorithms for use with imbalanced datasets, which commonly arise in many applications such as information extraction applications. Algorithms based on closest-to-hyperplane selection and query-by-committee selection are combined with methods ...
Support Vector Machine Active Learning Algorithms with Query-by-Committee versus Closest-to-Hyperplane Selection
2,018
http://arxiv.org/pdf/1801.07875v1
Title Support Vector Machine Active Learning Algorithms QuerybyCommittee versus ClosesttoHyperplane Selection Summary paper investigates evaluates support vector machine active learning algorithm use imbalanced datasets commonly arise many application information extraction application Algorithms based closesttohyperpl...
[0.04626673087477684, -0.04286425933241844, -0.03062080591917038, -0.019190559163689613, 0.011331303045153618, 0.029807595536112785, 6.210908759385347e-05, 0.030633825808763504, 0.025672173127532005, -0.0678013265132904, 0.02883731760084629, 0.00642591156065464, 0.029101606458425522, 0.022163646295666695, -0.0035041447...
1,874
1,874
['Lei Zhang', 'Shuai Wang', 'Bing Liu']
1801.07883v2
Deep learning has emerged as a powerful machine learning technique that learns multiple layers of representations or features of the data and produces state-of-the-art prediction results. Along with the success of deep learning in many other application domains, deep learning is also popularly used in sentiment analysi...
Deep Learning for Sentiment Analysis : A Survey
2,018
http://arxiv.org/pdf/1801.07883v2
Title Deep Learning Sentiment Analysis Survey Summary Deep learning emerged powerful machine learning technique learns multiple layer representation feature data produce stateoftheart prediction result Along success deep learning many application domain deep learning also popularly used sentiment analysis recent year p...
[1.4519059732265305e-06, 0.05628541484475136, -0.009255200624465942, 0.03389264643192291, -0.027464644983410835, 0.006450736429542303, -0.017552172765135765, -0.01115609984844923, 0.01880117878317833, -0.018571026623249054, 0.004483597818762064, 0.0028493539430201054, -0.029273107647895813, 0.05491868406534195, -0.0266...
1,875
1,875
['Garrett Beatty', 'Ethan Kochis', 'Michael Bloodgood']
1801.07887v1
When using active learning, smaller batch sizes are typically more efficient from a learning efficiency perspective. However, in practice due to speed and human annotator considerations, the use of larger batch sizes is necessary. While past work has shown that larger batch sizes decrease learning efficiency from a lea...
Impact of Batch Size on Stopping Active Learning for Text Classification
2,018
http://arxiv.org/pdf/1801.07887v1
Title Impact Batch Size Stopping Active Learning Text Classification Summary using active learning smaller batch size typically efficient learning efficiency perspective However practice due speed human annotator consideration use larger batch size necessary past work shown larger batch size decrease learning efficienc...
[0.02688148245215416, -0.02092062495648861, -0.01958298124372959, -0.02748035080730915, 0.00755170825868845, 0.040548309683799744, 0.03910109028220177, 0.05024091526865959, 0.024054130539298058, -0.0950501337647438, 0.022905806079506874, -0.0035278508439660072, -0.014161898754537106, 0.0507018119096756, -0.007649612147...
1,876
1,876
['Amitabha Karmakar']
1802.00382v1
We investigate the automatic classification of patient discharge notes into standard disease labels. We find that Convolutional Neural Networks with Attention outperform previous algorithms used in this task, and suggest further areas for improvement.
Classifying medical notes into standard disease codes using Machine Learning
2,018
http://arxiv.org/pdf/1802.00382v1
Title Classifying medical note standard disease code using Machine Learning Summary investigate automatic classification patient discharge note standard disease label find Convolutional Neural Networks Attention outperform previous algorithm used task suggest area improvement Authors 0 Ahmed Osman Wojciech Samek 1 Ji Y...
[0.046673864126205444, 0.025926979258656502, 0.009261058643460274, -0.00599827291443944, 0.01326112262904644, 0.03730321303009987, 0.04643900692462921, 0.03131059929728508, 0.0066565340384840965, -0.027061771601438522, 0.03682927042245865, -0.024745864793658257, 0.023209592327475548, 0.06615317612886429, 0.005178285762...
1,877
1,877
['Yoon Kim', 'Sam Wiseman', 'Andrew C. Miller', 'David Sontag', 'Alexander M. Rush']
1802.02550v3
Amortized variational inference (AVI) replaces instance-specific local inference with a global inference network. While AVI has enabled efficient training of deep generative models such as variational autoencoders (VAE), recent empirical work suggests that inference networks can produce suboptimal variational parameter...
Semi-Amortized Variational Autoencoders
2,018
http://arxiv.org/pdf/1802.02550v3
Title SemiAmortized Variational Autoencoders Summary Amortized variational inference AVI replaces instancespecific local inference global inference network AVI enabled efficient training deep generative model variational autoencoders VAE recent empirical work suggests inference network produce suboptimal variational pa...
[0.01025390811264515, 0.12230377644300461, -0.011734766885638237, 0.01654762029647827, 0.005407834891229868, 0.013828285038471222, 0.03790750727057457, -0.01663198508322239, -0.04452362284064293, 0.008540929295122623, 0.021560637280344963, -0.038954250514507294, 0.009988921694457531, 0.075185626745224, 0.05318577587604...
1,878
1,878
['Vlad Niculae', 'André F. T. Martins', 'Mathieu Blondel', 'Claire Cardie']
1802.04223v1
Structured prediction requires searching over a combinatorial number of structures. To tackle it, we introduce SparseMAP, a new method for sparse structured inference, together with corresponding loss functions. SparseMAP inference is able to automatically select only a few global structures: it is situated between MAP...
SparseMAP: Differentiable Sparse Structured Inference
2,018
http://arxiv.org/pdf/1802.04223v1
Title SparseMAP Differentiable Sparse Structured Inference Summary Structured prediction requires searching combinatorial number structure tackle introduce SparseMAP new method sparse structured inference together corresponding loss function SparseMAP inference able automatically select global structure situated MAP in...
[0.016037244349718094, 0.0801396369934082, 0.0031571213621646166, 0.062330462038517, -0.009647166356444359, -0.01477353647351265, -0.008576232939958572, 0.0030713393352925777, 0.01922258362174034, -0.03137041628360748, -0.015639912337064743, -0.01851065829396248, 0.013476704247295856, 0.057811543345451355, 0.0201306808...
1,879
1,879
['Xilun Chen', 'Claire Cardie']
1802.05694v1
Many text classification tasks are known to be highly domain-dependent. Unfortunately, the availability of training data can vary drastically across domains. Worse still, for some domains there may not be any annotated data at all. In this work, we propose a multinomial adversarial network (MAN) to tackle the text clas...
Multinomial Adversarial Networks for Multi-Domain Text Classification
2,018
http://arxiv.org/pdf/1802.05694v1
Title Multinomial Adversarial Networks MultiDomain Text Classification Summary Many text classification task known highly domaindependent Unfortunately availability training data vary drastically across domain Worse still domain may annotated data work propose multinomial adversarial network MAN tackle text classificat...
[0.0261891707777977, 0.029147107154130936, -0.011778626590967178, 0.03200707584619522, -0.031279414892196655, -0.002362688770517707, 0.06382735818624496, -0.001528458553366363, 0.012556253001093864, -0.06399880349636078, -0.059373531490564346, -0.03258289769291878, -0.008822670206427574, 0.019786208868026733, 0.0144760...
1,880
1,880
['James Mullenbach', 'Sarah Wiegreffe', 'Jon Duke', 'Jimeng Sun', 'Jacob Eisenstein']
1802.05695v1
Clinical notes are text documents that are created by clinicians for each patient encounter. They are typically accompanied by medical codes, which describe the diagnosis and treatment. Annotating these codes is labor intensive and error prone; furthermore, the connection between the codes and the text is not annotated...
Explainable Prediction of Medical Codes from Clinical Text
2,018
http://arxiv.org/pdf/1802.05695v1
Title Explainable Prediction Medical Codes Clinical Text Summary Clinical note text document created clinician patient encounter typically accompanied medical code describe diagnosis treatment Annotating code labor intensive error prone furthermore connection code text annotated obscuring reason detail behind specific ...
[0.014872699044644833, 0.04430777579545975, 0.017484629526734352, -0.010887043550610542, 0.02778918296098709, 0.043095435947179794, 0.002997482195496559, 0.037770967930555344, -0.0331706665456295, 0.027114642783999443, 0.030942102894186974, -0.032198693603277206, 0.029325250536203384, 0.08551100641489029, 0.02791857719...
1,881
1,881
['Fengyi Tang', 'Kaixiang Lin', 'Ikechukwu Uchendu', 'Hiroko H. Dodge', 'Jiayu Zhou']
1802.06428v1
Mild cognitive impairment (MCI) is a prodromal phase in the progression from normal aging to dementia, especially Alzheimers disease. Even though there is mild cognitive decline in MCI patients, they have normal overall cognition and thus is challenging to distinguish from normal aging. Using transcribed data obtained ...
Improving Mild Cognitive Impairment Prediction via Reinforcement Learning and Dialogue Simulation
2,018
http://arxiv.org/pdf/1802.06428v1
Title Improving Mild Cognitive Impairment Prediction via Reinforcement Learning Dialogue Simulation Summary Mild cognitive impairment MCI prodromal phase progression normal aging dementia especially Alzheimers disease Even though mild cognitive decline MCI patient normal overall cognition thus challenging distinguish n...
[0.029037868604063988, 0.03072071447968483, -0.023005638271570206, -0.011357119306921959, -0.018420139327645302, 0.017804095521569252, 0.012552016414701939, 0.02403733879327774, 0.05297079682350159, 0.015069079585373402, 0.02677818387746811, 0.015274985693395138, 0.03356941044330597, 0.02672848291695118, -0.03564117848...
1,882
1,882
['Kejun Huang', 'Xiao Fu', 'Nicholas D. Sidiropoulos']
1802.06894v1
We present a new algorithm for identifying the transition and emission probabilities of a hidden Markov model (HMM) from the emitted data. Expectation-maximization becomes computationally prohibitive for long observation records, which are often required for identification. The new algorithm is particularly suitable fo...
Learning Hidden Markov Models from Pairwise Co-occurrences with Applications to Topic Modeling
2,018
http://arxiv.org/pdf/1802.06894v1
Title Learning Hidden Markov Models Pairwise Cooccurrences Applications Topic Modeling Summary present new algorithm identifying transition emission probability hidden Markov model HMM emitted data Expectationmaximization becomes computationally prohibitive long observation record often required identification new algo...
[0.04771362617611885, 0.01918189227581024, 0.0082984808832407, 0.023779043927788734, -0.03001243807375431, 0.0030959455762058496, 0.030373159795999527, 0.026362257078289986, -0.04602859541773796, -0.06797562539577484, 0.017035409808158875, 0.02340453676879406, -0.015476190485060215, 0.05864096060395241, -0.039807919412...
1,883
1,883
['Jason Lee', 'Elman Mansimov', 'Kyunghyun Cho']
1802.06901v1
We propose a conditional non-autoregressive neural sequence model based on iterative refinement. The proposed model is designed based on the principles of latent variable models and denoising autoencoders, and is generally applicable to any sequence generation task. We extensively evaluate the proposed model on machine...
Deterministic Non-Autoregressive Neural Sequence Modeling by Iterative Refinement
2,018
http://arxiv.org/pdf/1802.06901v1
Title Deterministic NonAutoregressive Neural Sequence Modeling Iterative Refinement Summary propose conditional nonautoregressive neural sequence model based iterative refinement proposed model designed based principle latent variable model denoising autoencoders generally applicable sequence generation task extensivel...
[0.0204823799431324, 0.05376524105668068, -0.023124581202864647, 0.022667858749628067, -0.008469834923744202, -0.004963605664670467, 0.008173903450369835, -0.028385207056999207, -0.045095525681972504, -0.019197436049580574, 0.04989522695541382, -0.040421951562166214, 0.04335479810833931, 0.07870830595493317, 0.01446068...
1,884
1,884
['Ahmad Pesaranghader', 'Ali Pesaranghader', 'Stan Matwin', 'Marina Sokolova']
1802.09059v1
Due to recent technical and scientific advances, we have a wealth of information hidden in unstructured text data such as offline/online narratives, research articles, and clinical reports. To mine these data properly, attributable to their innate ambiguity, a Word Sense Disambiguation (WSD) algorithm can avoid numbers...
One Single Deep Bidirectional LSTM Network for Word Sense Disambiguation of Text Data
2,018
http://arxiv.org/pdf/1802.09059v1
Title One Single Deep Bidirectional LSTM Network Word Sense Disambiguation Text Data Summary Due recent technical scientific advance wealth information hidden unstructured text data offlineonline narrative research article clinical report mine data properly attributable innate ambiguity Word Sense Disambiguation WSD al...
[0.030580665916204453, -0.0012326767900958657, -0.004731867928057909, 0.06867516040802002, -0.027641814202070236, 0.009521962143480778, 0.009003321640193462, 0.02408500760793686, -0.001584772951900959, -0.04049181193113327, -0.004821751732379198, -0.051221489906311035, 0.010711639188230038, 0.06998465210199356, -0.0008...
1,885
1,885
['Niko Brummer', 'Anna Silnova', 'Lukas Burget', 'Themos Stafylakis']
1802.09777v1
Embeddings in machine learning are low-dimensional representations of complex input patterns, with the property that simple geometric operations like Euclidean distances and dot products can be used for classification and comparison tasks. The proposed meta-embeddings are special embeddings that live in more general in...
Gaussian meta-embeddings for efficient scoring of a heavy-tailed PLDA model
2,018
http://arxiv.org/pdf/1802.09777v1
Title Gaussian metaembeddings efficient scoring heavytailed PLDA model Summary Embeddings machine learning lowdimensional representation complex input pattern property simple geometric operation like Euclidean distance dot product used classification comparison task proposed metaembeddings special embeddings live gener...
[-0.01662541925907135, 0.013267363421618938, -0.005284935235977173, 0.07399339228868484, -0.007277723867446184, 0.0018509773071855307, 0.05727868527173996, -0.02431858889758587, -0.0621614046394825, -0.0015582650667056441, -0.015527958050370216, -0.008837644010782242, 0.035387955605983734, 0.0057166945189237595, 0.0187...
1,886
1,886
['Lifu Tu', 'Kevin Gimpel']
1803.03376v1
Structured prediction energy networks (SPENs; Belanger & McCallum 2016) use neural network architectures to define energy functions that can capture arbitrary dependencies among parts of structured outputs. Prior work used gradient descent for inference, relaxing the structured output to a set of continuous variables a...
Learning Approximate Inference Networks for Structured Prediction
2,018
http://arxiv.org/pdf/1803.03376v1
Title Learning Approximate Inference Networks Structured Prediction Summary Structured prediction energy network SPENs Belanger McCallum 2016 use neural network architecture define energy function capture arbitrary dependency among part structured output Prior work used gradient descent inference relaxing structured ou...
[0.001036197878420353, 0.04582258686423302, 0.002552865305915475, 0.028845099732279778, 0.01822827383875847, -0.006957597564905882, 0.010630166158080101, 0.015262567438185215, 0.005155600607395172, -0.03186600282788277, 0.016304755583405495, 0.028905214741826057, -0.007736696861684322, 0.04576975852251053, 0.0176914017...
1,887
1,887
['Ashish Vaswani', 'Samy Bengio', 'Eugene Brevdo', 'Francois Chollet', 'Aidan N. Gomez', 'Stephan Gouws', 'Llion Jones', 'Łukasz Kaiser', 'Nal Kalchbrenner', 'Niki Parmar', 'Ryan Sepassi', 'Noam Shazeer', 'Jakob Uszkoreit']
1803.07416v1
Tensor2Tensor is a library for deep learning models that is well-suited for neural machine translation and includes the reference implementation of the state-of-the-art Transformer model.
Tensor2Tensor for Neural Machine Translation
2,018
http://arxiv.org/pdf/1803.07416v1
Title Tensor2Tensor Neural Machine Translation Summary Tensor2Tensor library deep learning model wellsuited neural machine translation includes reference implementation stateoftheart Transformer model Authors 0 Ahmed Osman Wojciech Samek 1 Ji Young Lee Franck Dernoncourt 2 Iulian Vlad Serban Tim Klinger Gerald Tesau 3 ...
[0.024395586922764778, -0.0003874229732900858, 0.0027042936999350786, 0.03067980520427227, -0.024930991232395172, 0.01671147532761097, 0.04795849323272705, 0.02180713787674904, -0.04172877222299576, -0.01556894276291132, -0.006698892451822758, -0.04181475192308426, 0.05375088006258011, 0.022874128073453903, 0.031925439...
1,888
1,888
['Aurelia Bustos', 'Antonio Pertusa']
1803.08312v1
Interventional clinical cancer trials are generally too restrictive and cancer patients are often excluded from them on the basis of comorbidity, past or concomitant treatments and the fact that they are over a certain age. The efficacy and safety of new treatments for patients with these characteristics are not, there...
Learning Eligibility in Clinical Cancer Trials using Deep Neural Networks
2,018
http://arxiv.org/pdf/1803.08312v1
Title Learning Eligibility Clinical Cancer Trials using Deep Neural Networks Summary Interventional clinical cancer trial generally restrictive cancer patient often excluded basis comorbidity past concomitant treatment fact certain age efficacy safety new treatment patient characteristic therefore defined work build mo...
[0.04715457931160927, 0.06944018602371216, -0.0001923644740600139, -0.012957487255334854, -0.007507647387683392, 0.01569744385778904, 0.025114668533205986, 0.009190124459564686, -0.02146759442985058, -0.0036281407810747623, 0.011198149062693119, -0.045884281396865845, -0.016167404130101204, 0.0522306002676487, -0.01735...
1,889
1,889
['Yundong Zhang', 'Naveen Suda', 'Liangzhen Lai', 'Vikas Chandra']
1711.07128v3
Keyword spotting (KWS) is a critical component for enabling speech based user interactions on smart devices. It requires real-time response and high accuracy for good user experience. Recently, neural networks have become an attractive choice for KWS architecture because of their superior accuracy compared to tradition...
Hello Edge: Keyword Spotting on Microcontrollers
2,017
http://arxiv.org/pdf/1711.07128v3
Title Hello Edge Keyword Spotting Microcontrollers Summary Keyword spotting KWS critical component enabling speech based user interaction smart device requires realtime response high accuracy good user experience Recently neural network become attractive choice KWS architecture superior accuracy compared traditional sp...
[0.011477790772914886, 0.0013729825150221586, 0.017679890617728233, 0.08734149485826492, 0.025694970041513443, -0.04573433846235275, 0.02021219953894615, -0.006190213840454817, -0.02847370132803917, -0.03816450759768486, -0.02291201800107956, 0.0019354550167918205, 0.030789632350206375, 0.06268160790205002, 0.002454455...
1,890
1,890
['Andrej Karpathy', 'Armand Joulin', 'Li Fei-Fei']
1406.5679v1
We introduce a model for bidirectional retrieval of images and sentences through a multi-modal embedding of visual and natural language data. Unlike previous models that directly map images or sentences into a common embedding space, our model works on a finer level and embeds fragments of images (objects) and fragment...
Deep Fragment Embeddings for Bidirectional Image Sentence Mapping
2,014
http://arxiv.org/pdf/1406.5679v1
Title Deep Fragment Embeddings Bidirectional Image Sentence Mapping Summary introduce model bidirectional retrieval image sentence multimodal embedding visual natural language data Unlike previous model directly map image sentence common embedding space model work finer level embeds fragment image object fragment sente...
[0.03025973029434681, 0.03727172315120697, -0.013924416154623032, 0.09416542202234268, -0.041487645357847214, 0.00299708335660398, 0.003666093572974205, -0.015129479579627514, -0.01966891810297966, -0.05861286818981171, -0.0029089639429003, -0.027472978457808495, -0.03839787095785141, 0.07865127176046371, 0.03272298350...
1,891
1,891
['Haoyuan Gao', 'Junhua Mao', 'Jie Zhou', 'Zhiheng Huang', 'Lei Wang', 'Wei Xu']
1505.05612v3
In this paper, we present the mQA model, which is able to answer questions about the content of an image. The answer can be a sentence, a phrase or a single word. Our model contains four components: a Long Short-Term Memory (LSTM) to extract the question representation, a Convolutional Neural Network (CNN) to extract t...
Are You Talking to a Machine? Dataset and Methods for Multilingual Image Question Answering
2,015
http://arxiv.org/pdf/1505.05612v3
Title Talking Machine Dataset Methods Multilingual Image Question Answering Summary paper present mQA model able answer question content image answer sentence phrase single word model contains four component Long ShortTerm Memory LSTM extract question representation Convolutional Neural Network CNN extract visual repre...
[0.04367036744952202, 0.06498777866363525, 0.0006253236206248403, 0.056576069444417953, -0.003318394999951124, 0.01755686290562153, 0.03265567123889923, 0.0021714784670621157, -0.043752916157245636, -0.029157370328903198, -0.06535148620605469, -0.03422625735402107, 0.02712886780500412, 0.06963087618350983, 0.0292187910...
1,892
1,892
['Gordon Christie', 'Ankit Laddha', 'Aishwarya Agrawal', 'Stanislaw Antol', 'Yash Goyal', 'Kevin Kochersberger', 'Dhruv Batra']
1604.02125v4
We present an approach to simultaneously perform semantic segmentation and prepositional phrase attachment resolution for captioned images. Some ambiguities in language cannot be resolved without simultaneously reasoning about an associated image. If we consider the sentence "I shot an elephant in my pajamas", looking ...
Resolving Language and Vision Ambiguities Together: Joint Segmentation & Prepositional Attachment Resolution in Captioned Scenes
2,016
http://arxiv.org/pdf/1604.02125v4
Title Resolving Language Vision Ambiguities Together Joint Segmentation Prepositional Attachment Resolution Captioned Scenes Summary present approach simultaneously perform semantic segmentation prepositional phrase attachment resolution captioned image ambiguity language cannot resolved without simultaneously reasonin...
[0.05181458592414856, 0.05130207911133766, 0.02315017394721508, 0.03783247619867325, -0.033815860748291016, 0.029948312789201736, 0.02101781778037548, 0.03705848753452301, -0.013859746046364307, -0.05361881107091904, 0.0063454024493694305, 0.02064870297908783, 0.0372411273419857, 0.03835267201066017, -0.019517431035637...
1,893
1,893
['Álvaro Peris', 'Marc Bolaños', 'Petia Radeva', 'Francisco Casacuberta']
1604.03390v2
Although traditionally used in the machine translation field, the encoder-decoder framework has been recently applied for the generation of video and image descriptions. The combination of Convolutional and Recurrent Neural Networks in these models has proven to outperform the previous state of the art, obtaining more ...
Video Description using Bidirectional Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1604.03390v2
Title Video Description using Bidirectional Recurrent Neural Networks Summary Although traditionally used machine translation field encoderdecoder framework recently applied generation video image description combination Convolutional Recurrent Neural Networks model proven outperform previous state art obtaining accura...
[0.02621779963374138, 0.004300795961171389, 0.028350835666060448, 0.09264977276325226, -0.008883565664291382, -0.0020232731476426125, 0.00028703335556201637, -0.01378772221505642, -0.09416917711496353, -0.07014008611440659, 0.005109555087983608, -0.07925320416688919, 0.028285210952162743, 0.054809313267469406, 0.021628...
1,894
1,894
['Justin Johnson', 'Bharath Hariharan', 'Laurens van der Maaten', 'Li Fei-Fei', 'C. Lawrence Zitnick', 'Ross Girshick']
1612.06890v1
When building artificial intelligence systems that can reason and answer questions about visual data, we need diagnostic tests to analyze our progress and discover shortcomings. Existing benchmarks for visual question answering can help, but have strong biases that models can exploit to correctly answer questions witho...
CLEVR: A Diagnostic Dataset for Compositional Language and Elementary Visual Reasoning
2,016
http://arxiv.org/pdf/1612.06890v1
Title CLEVR Diagnostic Dataset Compositional Language Elementary Visual Reasoning Summary building artificial intelligence system reason answer question visual data need diagnostic test analyze progress discover shortcoming Existing benchmark visual question answering help strong bias model exploit correctly answer que...
[0.02765696682035923, 0.006148280575871468, -0.03456174209713936, 0.033993709832429886, -0.018513448536396027, 0.03092188574373722, 0.023070629686117172, 0.04360119625926018, 0.0007967096171341836, -0.019176891073584557, 0.042638108134269714, 0.027300292626023293, 0.025275088846683502, 0.06969999521970749, 0.0082039274...
1,895
1,895
['Gwangbeen Park', 'Woobin Im']
1612.08354v1
We present novel method for image-text multi-modal representation learning. In our knowledge, this work is the first approach of applying adversarial learning concept to multi-modal learning and not exploiting image-text pair information to learn multi-modal feature. We only use category information in contrast with mo...
Image-Text Multi-Modal Representation Learning by Adversarial Backpropagation
2,016
http://arxiv.org/pdf/1612.08354v1
Title ImageText MultiModal Representation Learning Adversarial Backpropagation Summary present novel method imagetext multimodal representation learning knowledge work first approach applying adversarial learning concept multimodal learning exploiting imagetext pair information learn multimodal feature use category inf...
[0.015570477582514286, 0.034042712301015854, -0.011000219732522964, 0.04251941666007042, 0.008999675512313843, -0.01811663992702961, 0.01752552017569542, 0.014520414173603058, -0.0008625760092400014, -0.06584911793470383, -0.031122924759984016, -0.004902757704257965, -0.01575721800327301, 0.007486236281692982, 0.065258...
1,896
1,896
['Matthew Ager', 'Zoran Cvetkovic', 'Peter Sollich']
1312.6849v2
Speech representation and modelling in high-dimensional spaces of acoustic waveforms, or a linear transformation thereof, is investigated with the aim of improving the robustness of automatic speech recognition to additive noise. The motivation behind this approach is twofold: (i) the information in acoustic waveforms ...
Speech Recognition Front End Without Information Loss
2,013
http://arxiv.org/pdf/1312.6849v2
Title Speech Recognition Front End Without Information Loss Summary Speech representation modelling highdimensional space acoustic waveform linear transformation thereof investigated aim improving robustness automatic speech recognition additive noise motivation behind approach twofold information acoustic waveform usu...
[-0.00012739536759909242, 0.03299427032470703, 0.01070835255086422, 0.03257551044225693, 0.027966486290097237, 0.0030043243896216154, 0.057797789573669434, 0.009790568612515926, -0.017805587500333786, -0.023766130208969116, -0.04008137807250023, -0.019896071404218674, 0.0626269280910492, 0.020837828516960144, 0.0151087...
1,897
1,897
['Junhua Mao', 'Wei Xu', 'Yi Yang', 'Jiang Wang', 'Alan L. Yuille']
1410.1090v1
In this paper, we present a multimodal Recurrent Neural Network (m-RNN) model for generating novel sentence descriptions to explain the content of images. It directly models the probability distribution of generating a word given previous words and the image. Image descriptions are generated by sampling from this distr...
Explain Images with Multimodal Recurrent Neural Networks
2,014
http://arxiv.org/pdf/1410.1090v1
Title Explain Images Multimodal Recurrent Neural Networks Summary paper present multimodal Recurrent Neural Network mRNN model generating novel sentence description explain content image directly model probability distribution generating word given previous word image Image description generated sampling distribution m...
[0.07071958482265472, 0.040514830499887466, -0.011414795182645321, 0.052854008972644806, -0.04709691181778908, -0.018606381490826607, 0.013258659280836582, -0.004676898010075092, -0.02842620387673378, -0.04802655428647995, 0.009444932453334332, -0.010040896013379097, -0.013090718537569046, 0.057345032691955566, 0.03341...
1,898
1,898
['Ryan Kiros', 'Ruslan Salakhutdinov', 'Richard S. Zemel']
1411.2539v1
Inspired by recent advances in multimodal learning and machine translation, we introduce an encoder-decoder pipeline that learns (a): a multimodal joint embedding space with images and text and (b): a novel language model for decoding distributed representations from our space. Our pipeline effectively unifies joint im...
Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models
2,014
http://arxiv.org/pdf/1411.2539v1
Title Unifying VisualSemantic Embeddings Multimodal Neural Language Models Summary Inspired recent advance multimodal learning machine translation introduce encoderdecoder pipeline learns multimodal joint embedding space image text b novel language model decoding distributed representation space pipeline effectively un...
[0.0261339470744133, 0.06662645190954208, 0.005605306476354599, 0.08023252338171005, -0.025240810588002205, 0.01120469719171524, 0.008748309686779976, 0.020405780524015427, -0.026412077248096466, -0.06169673055410385, -0.018997669219970703, -0.0030105954501777887, 0.019054794684052467, 0.056523244827985764, 0.043368414...
1,899
1,899
['Ha Jong Won', 'Li Gwang Chol', 'Kim Hyok Chol', 'Li Kum Song']
1411.4114v1
In this paper, we defined the viseme (visual speech element) and described about the method of extracting visual feature vector. We defined the 10 visemes based on vowel by analyzing of Korean utterance and proposed the method of extracting the 20-dimensional visual feature vector, combination of static features and dy...
Definition of Visual Speech Element and Research on a Method of Extracting Feature Vector for Korean Lip-Reading
2,014
http://arxiv.org/pdf/1411.4114v1
Title Definition Visual Speech Element Research Method Extracting Feature Vector Korean LipReading Summary paper defined viseme visual speech element described method extracting visual feature vector defined 10 visemes based vowel analyzing Korean utterance proposed method extracting 20dimensional visual feature vector...
[0.04889177903532982, -0.05082942545413971, 0.016762977465987206, 0.02013116143643856, 0.0018520386656746268, 0.018040856346488, 0.009472979232668877, 0.027332250028848648, 0.01199665293097496, 0.002173499669879675, -0.034919463098049164, 0.03333001211285591, 0.0851215347647667, 0.0272291898727417, 0.028246615082025528...