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
900
900
['Wei Wen', 'Chunpeng Wu', 'Yandan Wang', 'Yiran Chen', 'Hai Li']
1608.03665v4
High demand for computation resources severely hinders deployment of large-scale Deep Neural Networks (DNN) in resource constrained devices. In this work, we propose a Structured Sparsity Learning (SSL) method to regularize the structures (i.e., filters, channels, filter shapes, and layer depth) of DNNs. SSL can: (1) l...
Learning Structured Sparsity in Deep Neural Networks
2,016
http://arxiv.org/pdf/1608.03665v4
Title Learning Structured Sparsity Deep Neural Networks Summary High demand computation resource severely hinders deployment largescale Deep Neural Networks DNN resource constrained device work propose Structured Sparsity Learning SSL method regularize structure ie filter channel filter shape layer depth DNNs SSL 1 lea...
[-0.01659630611538887, 0.0748564675450325, -0.005367153324186802, 0.04604210704565048, 0.019997602328658104, -0.05583498626947403, 0.05846254155039787, 0.003320320276543498, -0.024876467883586884, 0.029903799295425415, 0.008502352982759476, 0.019730141386389732, 0.01199910044670105, 0.04032714664936066, -0.016805088147...
901
901
['Itay Safran', 'Ohad Shamir']
1610.09887v2
We provide several new depth-based separation results for feed-forward neural networks, proving that various types of simple and natural functions can be better approximated using deeper networks than shallower ones, even if the shallower networks are much larger. This includes indicators of balls and ellipses; non-lin...
Depth-Width Tradeoffs in Approximating Natural Functions with Neural Networks
2,016
http://arxiv.org/pdf/1610.09887v2
Title DepthWidth Tradeoffs Approximating Natural Functions Neural Networks Summary provide several new depthbased separation result feedforward neural network proving various type simple natural function better approximated using deeper network shallower one even shallower network much larger includes indicator ball el...
[-0.01175389438867569, 0.012245922349393368, -0.016391614452004433, 0.06159992888569832, -0.029291750863194466, -0.04699232801795006, 0.03928838297724724, -0.012378379702568054, -0.03245335817337036, 0.033251505345106125, 0.02554790861904621, -0.012750370427966118, 0.005581865087151527, 0.035941287875175476, 0.01612184...
902
902
['Chuan-Yung Tsai', 'Andrew Saxe', 'David Cox']
1610.10087v1
We present a novel neural network algorithm, the Tensor Switching (TS) network, which generalizes the Rectified Linear Unit (ReLU) nonlinearity to tensor-valued hidden units. The TS network copies its entire input vector to different locations in an expanded representation, with the location determined by its hidden un...
Tensor Switching Networks
2,016
http://arxiv.org/pdf/1610.10087v1
Title Tensor Switching Networks Summary present novel neural network algorithm Tensor Switching TS network generalizes Rectified Linear Unit ReLU nonlinearity tensorvalued hidden unit TS network copy entire input vector different location expanded representation location determined hidden unit activity way even simple ...
[-0.038573481142520905, -0.017879076302051544, -0.022742094472050667, 0.04812457785010338, -0.006141091696918011, 0.012086226604878902, 0.06394513696432114, 0.01881677284836769, -0.007223112974315882, -0.029840851202607155, -0.024261638522148132, 0.0003172199649270624, -0.013522958382964134, 0.04881155118346214, 0.0088...
903
903
['Maithra Raghu', 'Ben Poole', 'Jon Kleinberg', 'Surya Ganguli', 'Jascha Sohl-Dickstein']
1611.08083v1
We survey results on neural network expressivity described in "On the Expressive Power of Deep Neural Networks". The paper motivates and develops three natural measures of expressiveness, which all display an exponential dependence on the depth of the network. In fact, all of these measures are related to a fourth quan...
Survey of Expressivity in Deep Neural Networks
2,016
http://arxiv.org/pdf/1611.08083v1
Title Survey Expressivity Deep Neural Networks Summary survey result neural network expressivity described Expressive Power Deep Neural Networks paper motivates develops three natural measure expressiveness display exponential dependence depth network fact measure related fourth quantity trajectory length quantity grow...
[-0.040803201496601105, 0.0421065017580986, -0.038796842098236084, 0.017388811334967613, -0.01189966220408678, -0.061782948672771454, 0.018624713644385338, 0.013946075923740864, -0.07145845144987106, 0.03328273817896843, 0.00952074769884348, 0.008740254677832127, 0.01378615852445364, 0.08721750974655151, 0.025093082338...
904
904
['Zachary C. Lipton', 'Subarna Tripathi']
1702.04782v2
Generative adversarial networks (GANs) transform latent vectors into visually plausible images. It is generally thought that the original GAN formulation gives no out-of-the-box method to reverse the mapping, projecting images back into latent space. We introduce a simple, gradient-based technique called stochastic cli...
Precise Recovery of Latent Vectors from Generative Adversarial Networks
2,017
http://arxiv.org/pdf/1702.04782v2
Title Precise Recovery Latent Vectors Generative Adversarial Networks Summary Generative adversarial network GANs transform latent vector visually plausible image generally thought original GAN formulation give outofthebox method reverse mapping projecting image back latent space introduce simple gradientbased techniqu...
[0.026400696486234665, 0.07534674555063248, -0.01070816908031702, 0.05447891727089882, -0.026032110676169395, -0.03902760148048401, -0.007082941010594368, 0.019612129777669907, -0.09977174550294876, 0.03497481718659401, 0.01035909540951252, 0.014261354692280293, -0.007608470041304827, 0.02977430634200573, 0.07887084782...
905
905
['Nathan Ng', 'Rodney A Gabriel', 'Julian McAuley', 'Charles Elkan', 'Zachary C Lipton']
1702.05386v3
Scheduling surgeries is a challenging task due to the fundamental uncertainty of the clinical environment, as well as the risks and costs associated with under- and over-booking. We investigate neural regression algorithms to estimate the parameters of surgery case durations, focusing on the issue of heteroscedasticity...
Predicting Surgery Duration with Neural Heteroscedastic Regression
2,017
http://arxiv.org/pdf/1702.05386v3
Title Predicting Surgery Duration Neural Heteroscedastic Regression Summary Scheduling surgery challenging task due fundamental uncertainty clinical environment well risk cost associated overbooking investigate neural regression algorithm estimate parameter surgery case duration focusing issue heteroscedasticity seek s...
[0.0035067747812718153, 0.05280238389968872, -0.05167202278971672, -0.06231819838285446, 0.026894744485616684, 0.014329604804515839, 0.04932184889912605, 0.020540937781333923, -0.057872798293828964, 0.02975635416805744, 0.020651469007134438, -0.047121625393629074, 0.024571049958467484, 0.08178380876779556, 0.0296381805...
906
906
['Haihao Lu', 'Kenji Kawaguchi']
1702.08580v2
In deep learning, \textit{depth}, as well as \textit{nonlinearity}, create non-convex loss surfaces. Then, does depth alone create bad local minima? In this paper, we prove that without nonlinearity, depth alone does not create bad local minima, although it induces non-convex loss surface. Using this insight, we greatl...
Depth Creates No Bad Local Minima
2,017
http://arxiv.org/pdf/1702.08580v2
Title Depth Creates Bad Local Minima Summary deep learning textitdepth well textitnonlinearity create nonconvex loss surface depth alone create bad local minimum paper prove without nonlinearity depth alone create bad local minimum although induces nonconvex loss surface Using insight greatly simplify recently proposed...
[-0.017659468576312065, 0.04269912838935852, 0.004770276136696339, 0.04584232717752457, -0.03638337180018425, -0.019156763330101967, 0.02794671803712845, 0.010106617584824562, -0.04143384471535683, 0.022138074040412903, 0.004321425687521696, 0.02169121615588665, 0.01978026144206524, 0.053596775978803635, 0.048114225268...
907
907
['Kenji Kawaguchi', 'Bo Xie', 'Vikas Verma', 'Le Song']
1702.08882v7
We propose semi-random features for nonlinear function approximation. The flexibility of semi-random feature lies between the fully adjustable units in deep learning and the random features used in kernel methods. For one hidden layer models with semi-random features, we prove with no unrealistic assumptions that the m...
Deep Semi-Random Features for Nonlinear Function Approximation
2,017
http://arxiv.org/pdf/1702.08882v7
Title Deep SemiRandom Features Nonlinear Function Approximation Summary propose semirandom feature nonlinear function approximation flexibility semirandom feature lie fully adjustable unit deep learning random feature used kernel method one hidden layer model semirandom feature prove unrealistic assumption model class ...
[-0.03810715675354004, 0.042253658175468445, -0.016475047916173935, 0.03775469586253166, 0.02621050551533699, -0.038027409464120865, 0.054141078144311905, -0.0273752398788929, -0.04127400740981102, 0.000567191222216934, -0.010257388465106487, -0.03006933629512787, -0.03344865143299103, 0.07191070169210434, 0.0183857977...
908
908
['Pietro Morerio', 'Jacopo Cavazza', 'Riccardo Volpi', 'Rene Vidal', 'Vittorio Murino']
1703.06229v2
Dropout is a very effective way of regularizing neural networks. Stochastically "dropping out" units with a certain probability discourages over-specific co-adaptations of feature detectors, preventing overfitting and improving network generalization. Besides, Dropout can be interpreted as an approximate model aggregat...
Curriculum Dropout
2,017
http://arxiv.org/pdf/1703.06229v2
Title Curriculum Dropout Summary Dropout effective way regularizing neural network Stochastically dropping unit certain probability discourages overspecific coadaptations feature detector preventing overfitting improving network generalization Besides Dropout interpreted approximate model aggregation technique exponent...
[0.021196754649281502, 0.025057265534996986, -0.04824516177177429, 0.02435699664056301, 0.0032089315354824066, 0.01290759164839983, 0.05139574781060219, -0.03810380399227142, -0.018367275595664978, 0.004343180451542139, -0.004008772782981396, 0.051786135882139206, 0.005155432969331741, 0.04700867086648941, 0.0343609675...
909
909
['David Rolnick', 'Max Tegmark']
1705.05502v1
It is well-known that neural networks are universal approximators, but that deeper networks tend to be much more efficient than shallow ones. We shed light on this by proving that the total number of neurons $m$ required to approximate natural classes of multivariate polynomials of $n$ variables grows only linearly wit...
The power of deeper networks for expressing natural functions
2,017
http://arxiv.org/pdf/1705.05502v1
Title power deeper network expressing natural function Summary wellknown neural network universal approximators deeper network tend much efficient shallow one shed light proving total number neuron required approximate natural class multivariate polynomial n variable grows linearly n deep neural network grows exponenti...
[-0.06329913437366486, 0.08246143907308578, -0.015012403950095177, 0.019937817007303238, -0.01387228537350893, -0.04648955538868904, 0.02122058905661106, -0.03758176788687706, -0.010652941651642323, 0.02578652650117874, 0.04564613103866577, 0.02698303945362568, -0.020929984748363495, 0.042022790759801865, 0.07929632812...
910
910
['Dongsung Huh', 'Terrence J. Sejnowski']
1706.04698v2
Much of studies on neural computation are based on network models of static neurons that produce analog output, despite the fact that information processing in the brain is predominantly carried out by dynamic neurons that produce discrete pulses called spikes. Research in spike-based computation has been impeded by th...
Gradient Descent for Spiking Neural Networks
2,017
http://arxiv.org/pdf/1706.04698v2
Title Gradient Descent Spiking Neural Networks Summary Much study neural computation based network model static neuron produce analog output despite fact information processing brain predominantly carried dynamic neuron produce discrete pulse called spike Research spikebased computation impeded lack efficient supervise...
[-0.030215846374630928, -0.061905812472105026, -0.020115753635764122, 0.03791351616382599, 0.030119411647319794, -0.001504110754467547, 0.002473750151693821, -0.004003675188869238, 0.02099120244383812, 0.017014838755130768, -0.024780504405498505, 0.03629687428474426, -0.00887641217559576, 0.061002928763628006, 0.011684...
911
911
['Ivo Gonçalves', 'Sara Silva', 'Carlos M. Fonseca', 'Mauro Castelli']
1706.06195v1
In iterative supervised learning algorithms it is common to reach a point in the search where no further induction seems to be possible with the available data. If the search is continued beyond this point, the risk of overfitting increases significantly. Following the recent developments in inductive semantic stochast...
Unsure When to Stop? Ask Your Semantic Neighbors
2,017
http://arxiv.org/pdf/1706.06195v1
Title Unsure Stop Ask Semantic Neighbors Summary iterative supervised learning algorithm common reach point search induction seems possible available data search continued beyond point risk overfitting increase significantly Following recent development inductive semantic stochastic method paper study feasibility using...
[0.0018006833270192146, 0.025668496266007423, -0.01144926343113184, 0.013653518632054329, -0.029787158593535423, 0.004409995395690203, -0.01718537136912346, 0.00735758850350976, -0.022570738568902016, 0.009048466570675373, 0.04725297540426254, 0.0631190687417984, -0.00830131582915783, -0.013563445769250393, -0.00094968...
912
912
['Daniel Hsu']
1708.02975v2
In this paper, we use variational recurrent neural network to investigate the anomaly detection problem on graph time series. The temporal correlation is modeled by the combination of recurrent neural network (RNN) and variational inference (VI), while the spatial information is captured by the graph convolutional netw...
Anomaly Detection on Graph Time Series
2,017
http://arxiv.org/pdf/1708.02975v2
Title Anomaly Detection Graph Time Series Summary paper use variational recurrent neural network investigate anomaly detection problem graph time series temporal correlation modeled combination recurrent neural network RNN variational inference VI spatial information captured graph convolutional network order incorpora...
[0.0027484481688588858, 0.03369508311152458, -0.02338779717683792, 0.02868322655558586, -0.08980365097522736, -0.015192878432571888, 0.010118133388459682, -0.014228387735784054, -0.00973463337868452, 0.0182920191437006, 0.10572895407676697, 0.020029330626130104, -0.012014509178698063, 0.10851632058620453, 0.00167819787...
913
913
['Abien Fred Agarap']
1709.03082v7
Gated Recurrent Unit (GRU) is a recently-developed variation of the long short-term memory (LSTM) unit, both of which are types of recurrent neural network (RNN). Through empirical evidence, both models have been proven to be effective in a wide variety of machine learning tasks such as natural language processing (Wen...
A Neural Network Architecture Combining Gated Recurrent Unit (GRU) and Support Vector Machine (SVM) for Intrusion Detection in Network Traffic Data
2,017
http://arxiv.org/pdf/1709.03082v7
Title Neural Network Architecture Combining Gated Recurrent Unit GRU Support Vector Machine SVM Intrusion Detection Network Traffic Data Summary Gated Recurrent Unit GRU recentlydeveloped variation long shortterm memory LSTM unit type recurrent neural network RNN empirical evidence model proven effective wide variety m...
[0.038477201014757156, -0.022839749231934547, -0.019205721095204353, 0.04534827545285225, -0.05568360164761543, -0.02706184610724449, 0.06314827501773834, -0.037587426602840424, 0.027869094163179398, -0.06569738686084747, 0.04174518957734108, -0.03659254312515259, -0.011877326294779778, 0.11472046375274658, 0.021196575...
914
914
['Divya Gopinath', 'Guy Katz', 'Corina S. Pasareanu', 'Clark Barrett']
1710.00486v1
Deep neural networks have become widely used, obtaining remarkable results in domains such as computer vision, speech recognition, natural language processing, audio recognition, social network filtering, machine translation, and bio-informatics, where they have produced results comparable to human experts. However, th...
DeepSafe: A Data-driven Approach for Checking Adversarial Robustness in Neural Networks
2,017
http://arxiv.org/pdf/1710.00486v1
Title DeepSafe Datadriven Approach Checking Adversarial Robustness Neural Networks Summary Deep neural network become widely used obtaining remarkable result domain computer vision speech recognition natural language processing audio recognition social network filtering machine translation bioinformatics produced resul...
[-0.01810164377093315, 0.044370271265506744, -0.03122018091380596, 0.03759416937828064, -0.0023366259410977364, -0.043730318546295166, 0.043661925941705704, -0.03650018945336342, 0.0038855327293276787, -0.005678831599652767, 0.012616811320185661, 0.017511045560240746, 0.013287299312651157, 0.05987581983208656, 0.068602...
915
915
['Grzegorz Dudek']
1710.04874v1
Neural networks with random hidden nodes have gained increasing interest from researchers and practical applications. This is due to their unique features such as very fast training and universal approximation property. In these networks the weights and biases of hidden nodes determining the nonlinear feature mapping a...
A Method of Generating Random Weights and Biases in Feedforward Neural Networks with Random Hidden Nodes
2,017
http://arxiv.org/pdf/1710.04874v1
Title Method Generating Random Weights Biases Feedforward Neural Networks Random Hidden Nodes Summary Neural network random hidden node gained increasing interest researcher practical application due unique feature fast training universal approximation property network weight bias hidden node determining nonlinear feat...
[-0.0025384598411619663, -0.0127226822078228, -0.03115057945251465, 0.00018343626288697124, -0.008374815806746483, -0.04033152014017105, 0.04117656871676445, -0.032881516963243484, 0.04597532004117966, 0.014851347543299198, 0.0009996519656851888, 0.0401940755546093, 0.011621665209531784, 0.03633009269833565, 0.03180719...
916
916
['Rumen Dangovski', 'Li Jing', 'Marin Soljacic']
1710.09537v1
The concepts of unitary evolution matrices and associative memory have boosted the field of Recurrent Neural Networks (RNN) to state-of-the-art performance in a variety of sequential tasks. However, RNN still have a limited capacity to manipulate long-term memory. To bypass this weakness the most successful application...
Rotational Unit of Memory
2,017
http://arxiv.org/pdf/1710.09537v1
Title Rotational Unit Memory Summary concept unitary evolution matrix associative memory boosted field Recurrent Neural Networks RNN stateoftheart performance variety sequential task However RNN still limited capacity manipulate longterm memory bypass weakness successful application RNN use external technique attention...
[0.03185699135065079, 0.025411365553736687, -0.014463833533227444, 0.04377507418394089, -0.008871356956660748, 0.031174534931778908, 0.005433708429336548, 0.0005734682781621814, -0.06515191495418549, 0.0006912930984981358, -0.03514276072382927, -0.06120593100786209, 0.011213659308850765, -0.00538595812395215, 0.0273787...
917
917
['Tero Karras', 'Timo Aila', 'Samuli Laine', 'Jaakko Lehtinen']
1710.10196v3
We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes i...
Progressive Growing of GANs for Improved Quality, Stability, and Variation
2,017
http://arxiv.org/pdf/1710.10196v3
Title Progressive Growing GANs Improved Quality Stability Variation Summary describe new training methodology generative adversarial network key idea grow generator discriminator progressively starting low resolution add new layer model increasingly fine detail training progress speed training greatly stabilizes allowi...
[0.01871155947446823, 0.11053736507892609, -0.011309931986033916, 0.024971190840005875, 0.017558861523866653, -0.028626861050724983, 0.015793029218912125, 0.004993888549506664, -0.03921636939048767, 0.04815524071455002, 0.029475171118974686, 0.0026604984886944294, -0.02696581743657589, 0.022566430270671844, 0.054071109...
918
918
['Cem Subakan', 'Paris Smaragdis']
1710.10779v1
Generative source separation methods such as non-negative matrix factorization (NMF) or auto-encoders, rely on the assumption of an output probability density. Generative Adversarial Networks (GANs) can learn data distributions without needing a parametric assumption on the output density. We show on a speech source se...
Generative Adversarial Source Separation
2,017
http://arxiv.org/pdf/1710.10779v1
Title Generative Adversarial Source Separation Summary Generative source separation method nonnegative matrix factorization NMF autoencoders rely assumption output probability density Generative Adversarial Networks GANs learn data distribution without needing parametric assumption output density show speech source sep...
[-0.00513956043869257, 0.08517022430896759, 0.006246505305171013, 0.030120765790343285, 0.0018776756478473544, -0.02955840528011322, 0.027502423152327538, -0.0003878129937220365, -0.09014207869768143, -0.00039256407762877643, -0.0823880136013031, -0.006068127229809761, -0.013439353555440903, 0.0016103184316307306, 0.06...
919
919
['Yuan Zeng', 'Kevin Devincentis', 'Yao Xiao', 'Zubayer Ibne Ferdous', 'Xiaochen Guo', 'Zhiyuan Yan', 'Yevgeny Berdichevsky']
1710.10944v3
Neural networks have shown great potential in many applications like speech recognition, drug discovery, image classification, and object detection. Neural network models are inspired by biological neural networks, but they are optimized to perform machine learning tasks on digital computers. The proposed work explores...
A Supervised STDP-based Training Algorithm for Living Neural Networks
2,017
http://arxiv.org/pdf/1710.10944v3
Title Supervised STDPbased Training Algorithm Living Neural Networks Summary Neural network shown great potential many application like speech recognition drug discovery image classification object detection Neural network model inspired biological neural network optimized perform machine learning task digital computer...
[-0.012166383676230907, -0.008391223847866058, -0.018980277702212334, 0.04643481224775314, 0.027766350656747818, 3.060038943658583e-05, 0.059188112616539, -0.008656064979732037, 0.01536852028220892, 0.006702076178044081, -0.016468089073896408, 0.018291722983121872, 0.021369101479649544, 0.04788945987820625, 0.043786711...
920
920
['John Alberg', 'Zachary C. Lipton']
1711.04837v1
On a periodic basis, publicly traded companies are required to report fundamentals: financial data such as revenue, operating income, debt, among others. These data points provide some insight into the financial health of a company. Academic research has identified some factors, i.e. computed features of the reported d...
Improving Factor-Based Quantitative Investing by Forecasting Company Fundamentals
2,017
http://arxiv.org/pdf/1711.04837v1
Title Improving FactorBased Quantitative Investing Forecasting Company Fundamentals Summary periodic basis publicly traded company required report fundamental financial data revenue operating income debt among others data point provide insight financial health company Academic research identified factor ie computed fea...
[-0.02115362137556076, 0.06896302849054337, -0.037331871688365936, -0.028223684057593346, 0.021599896252155304, -0.04645337536931038, 0.004053597804158926, 0.03677145391702652, -0.06326203048229218, -0.024846667423844337, 0.015410063788294792, 0.04602045193314552, 0.0012376694940030575, 0.14129263162612915, 0.012704470...
921
921
['Eli David', 'Moshe Koppel', 'Nathan S. Netanyahu']
1711.06839v1
In this paper we demonstrate how genetic algorithms can be used to reverse engineer an evaluation function's parameters for computer chess. Our results show that using an appropriate mentor, we can evolve a program that is on par with top tournament-playing chess programs, outperforming a two-time World Computer Chess ...
Genetic Algorithms for Mentor-Assisted Evaluation Function Optimization
2,017
http://arxiv.org/pdf/1711.06839v1
Title Genetic Algorithms MentorAssisted Evaluation Function Optimization Summary paper demonstrate genetic algorithm used reverse engineer evaluation function parameter computer chess result show using appropriate mentor evolve program par top tournamentplaying chess program outperforming twotime World Computer Chess C...
[0.027218738570809364, 0.04186451435089111, -0.029778560623526573, 0.042456574738025665, -0.016284598037600517, -0.03932826593518257, -0.05600321292877197, 0.030629165470600128, 0.008055981248617172, -0.024177247658371925, 0.03700937330722809, -0.008535700850188732, 0.0029371825512498617, 0.02641790732741356, 0.0218542...
922
922
['Guglielmo Montone', "J. Kevin O'Regan", 'Alexander V. Terekhov']
1711.10204v1
In the present work we propose a Deep Feed Forward network architecture which can be trained according to a sequential learning paradigm, where tasks of increasing difficulty are learned sequentially, yet avoiding catastrophic forgetting. The proposed architecture can re-use the features learned on previous tasks in a ...
Block Neural Network Avoids Catastrophic Forgetting When Learning Multiple Task
2,017
http://arxiv.org/pdf/1711.10204v1
Title Block Neural Network Avoids Catastrophic Forgetting Learning Multiple Task Summary present work propose Deep Feed Forward network architecture trained according sequential learning paradigm task increasing difficulty learned sequentially yet avoiding catastrophic forgetting proposed architecture reuse feature lea...
[-0.024706732481718063, 0.03879423812031746, -0.018889041617512703, -0.0029749220702797174, 0.010376326739788055, 0.020491337403655052, 0.03496556729078293, -0.034987542778253555, 0.01740151271224022, 0.04554947465658188, -0.012515983544290066, -0.02676737867295742, -0.023947542533278465, 0.045896418392658234, 0.033725...
923
923
['Kyeong Soo Kim', 'Sanghyuk Lee', 'Kaizhu Huang']
1712.01990v1
One of the key technologies for future large-scale location-aware services covering a complex of multi-story buildings --- e.g., a big shopping mall and a university campus --- is a scalable indoor localization technique. In this paper, we report the current status of our investigation on the use of deep neural network...
A Scalable Deep Neural Network Architecture for Multi-Building and Multi-Floor Indoor Localization Based on Wi-Fi Fingerprinting
2,017
http://arxiv.org/pdf/1712.01990v1
Title Scalable Deep Neural Network Architecture MultiBuilding MultiFloor Indoor Localization Based WiFi Fingerprinting Summary One key technology future largescale locationaware service covering complex multistory building eg big shopping mall university campus scalable indoor localization technique paper report curren...
[-0.04584827646613121, 0.0041849929839372635, -0.018235113471746445, -0.02328268066048622, -0.027477629482746124, 0.007399750407785177, 0.06161508336663246, -0.06816114485263824, -0.027744336053729057, -0.017726950347423553, -0.046600449830293655, 0.032152097672224045, 0.026632782071828842, 0.028290538117289543, 0.0143...
924
924
['Rudy Raymond', 'Takayuki Osogami', 'Sakyasingha Dasgupta']
1712.06132v1
Dynamic Boltzmann Machine (DyBM) has been shown highly efficient to predict time-series data. Gaussian DyBM is a DyBM that assumes the predicted data is generated by a Gaussian distribution whose first-order moment (mean) dynamically changes over time but its second-order moment (variance) is fixed. However, in many fi...
Dynamic Boltzmann Machines for Second Order Moments and Generalized Gaussian Distributions
2,017
http://arxiv.org/pdf/1712.06132v1
Title Dynamic Boltzmann Machines Second Order Moments Generalized Gaussian Distributions Summary Dynamic Boltzmann Machine DyBM shown highly efficient predict timeseries data Gaussian DyBM DyBM assumes predicted data generated Gaussian distribution whose firstorder moment mean dynamically change time secondorder moment...
[-0.005392593331634998, -0.030938461422920227, -0.056202471256256104, -0.04896299168467522, 0.027759017422795296, 0.006648993585258722, 0.010414808988571167, -0.0027628049720078707, -0.05353260412812233, 0.05078194662928581, 0.012549011036753654, 0.005730956327170134, 0.0038401647470891476, 0.07450827211141586, 0.02484...
925
925
['Marco Martinolli', 'Wulfram Gerstner', 'Aditya Gilra']
1712.10062v1
Learning and memory are intertwined in our brain and their relationship is at the core of several recent neural network models. In particular, the Attention-Gated MEmory Tagging model (AuGMEnT) is a reinforcement learning network with an emphasis on biological plausibility of memory dynamics and learning. We find that ...
Multi-timescale memory dynamics in a reinforcement learning network with attention-gated memory
2,017
http://arxiv.org/pdf/1712.10062v1
Title Multitimescale memory dynamic reinforcement learning network attentiongated memory Summary Learning memory intertwined brain relationship core several recent neural network model particular AttentionGated MEmory Tagging model AuGMEnT reinforcement learning network emphasis biological plausibility memory dynamic l...
[0.011148829013109207, -0.04073084890842438, -0.03570118173956871, -0.023461485281586647, 0.00968702882528305, 0.028548099100589752, 0.022292574867606163, -0.031344763934612274, -0.02595648169517517, 0.0017757597379386425, -0.03176902234554291, -0.013322173617780209, 0.020687390118837357, 0.05467847362160683, 0.0242297...
926
926
['Enrique Romero Merino', 'Ferran Mazzanti Castrillejo', 'Jordi Delgado Pin', 'David Buchaca Prats']
1801.02567v1
Learning algorithms for energy based Boltzmann architectures that rely on gradient descent are in general computationally prohibitive, typically due to the exponential number of terms involved in computing the partition function. In this way one has to resort to approximation schemes for the evaluation of the gradient....
Weighted Contrastive Divergence
2,018
http://arxiv.org/pdf/1801.02567v1
Title Weighted Contrastive Divergence Summary Learning algorithm energy based Boltzmann architecture rely gradient descent general computationally prohibitive typically due exponential number term involved computing partition function way one resort approximation scheme evaluation gradient case Restricted Boltzmann Mac...
[-4.9399870476918295e-05, -0.028885919600725174, -0.039195675402879715, 0.03501042351126671, -0.023319285362958908, 0.0015925582265481353, 0.04114723578095436, 0.0066097634844481945, -0.007172462530434132, 0.04707513377070427, 0.004055690951645374, -0.019096169620752335, 0.020037123933434486, 0.034189775586128235, 0.03...
927
927
['Shinichi Shirakawa', 'Yasushi Iwata', 'Youhei Akimoto']
1801.07650v1
Deep neural networks (DNNs) are powerful machine learning models and have succeeded in various artificial intelligence tasks. Although various architectures and modules for the DNNs have been proposed, selecting and designing the appropriate network structure for a target problem is a challenging task. In this paper, w...
Dynamic Optimization of Neural Network Structures Using Probabilistic Modeling
2,018
http://arxiv.org/pdf/1801.07650v1
Title Dynamic Optimization Neural Network Structures Using Probabilistic Modeling Summary Deep neural network DNNs powerful machine learning model succeeded various artificial intelligence task Although various architecture module DNNs proposed selecting designing appropriate network structure target problem challengin...
[-0.013948259875178337, 0.015110231004655361, -0.04892431199550629, 0.01586272194981575, -0.007155288942158222, -0.05642760545015335, 0.026939455419778824, -0.04784902185201645, -0.03239194676280022, 0.016747971996665, 0.0051847235299646854, -0.01960200071334839, 0.03455367684364319, 0.06773195415735245, 0.030500292778...
928
928
['Mauro Castelli', 'Ivo Gonçalves', 'Luca Manzoni', 'Leonardo Vanneschi']
1801.07668v1
The objective of this paper is to define an effective strategy for building an ensemble of Genetic Programming (GP) models. Ensemble methods are widely used in machine learning due to their features: they average out biases, they reduce the variance and they usually generalize better than single models. Despite these a...
Pruning Techniques for Mixed Ensembles of Genetic Programming Models
2,018
http://arxiv.org/pdf/1801.07668v1
Title Pruning Techniques Mixed Ensembles Genetic Programming Models Summary objective paper define effective strategy building ensemble Genetic Programming GP model Ensemble method widely used machine learning due feature average bias reduce variance usually generalize better single model Despite advantage building ens...
[-0.015443344600498676, 0.06426004320383072, -0.04797778278589249, 0.003623044118285179, -0.04073004424571991, 0.008288819342851639, -0.02020888403058052, 0.04328633099794388, -0.024020640179514885, -0.0010451363632455468, 0.0330679826438427, -0.014003001153469086, 0.032348960638046265, 0.02480197325348854, -0.05285409...
929
929
['Guillaume Desjardins', 'Razvan Pascanu', 'Aaron Courville', 'Yoshua Bengio']
1301.3545v2
This paper introduces the Metric-Free Natural Gradient (MFNG) algorithm for training Boltzmann Machines. Similar in spirit to the Hessian-Free method of Martens [8], our algorithm belongs to the family of truncated Newton methods and exploits an efficient matrix-vector product to avoid explicitely storing the natural g...
Metric-Free Natural Gradient for Joint-Training of Boltzmann Machines
2,013
http://arxiv.org/pdf/1301.3545v2
Title MetricFree Natural Gradient JointTraining Boltzmann Machines Summary paper introduces MetricFree Natural Gradient MFNG algorithm training Boltzmann Machines Similar spirit HessianFree method Martens 8 algorithm belongs family truncated Newton method exploit efficient matrixvector product avoid explicitely storing...
[-0.03228992596268654, -0.0158121008425951, 0.0068503329530358315, 0.05876896157860756, 0.01943853870034218, 0.019549576565623283, 0.037552040070295334, 0.00020413470338098705, -0.012966914102435112, 0.010233981534838676, -0.04960251972079277, 0.011127850972115993, -0.013932737521827221, 0.01020741555839777, 0.02560720...
930
930
['Matthew D. Zeiler', 'Rob Fergus']
1301.3557v1
We introduce a simple and effective method for regularizing large convolutional neural networks. We replace the conventional deterministic pooling operations with a stochastic procedure, randomly picking the activation within each pooling region according to a multinomial distribution, given by the activities within th...
Stochastic Pooling for Regularization of Deep Convolutional Neural Networks
2,013
http://arxiv.org/pdf/1301.3557v1
Title Stochastic Pooling Regularization Deep Convolutional Neural Networks Summary introduce simple effective method regularizing large convolutional neural network replace conventional deterministic pooling operation stochastic procedure randomly picking activation within pooling region according multinomial distribut...
[-0.0109257148578763, 0.06264904886484146, 0.0043680788949131966, 0.020792240276932716, 0.015698611736297607, -0.01729300431907177, 0.046514227986335754, -0.013023522682487965, -0.04297305643558502, 0.01335382554680109, -0.007356017362326384, -0.006719722878187895, 0.021091314032673836, 0.015134235844016075, 0.03541594...
931
931
['Ryan Kiros']
1301.3641v3
Hessian-free (HF) optimization has been successfully used for training deep autoencoders and recurrent networks. HF uses the conjugate gradient algorithm to construct update directions through curvature-vector products that can be computed on the same order of time as gradients. In this paper we exploit this property a...
Training Neural Networks with Stochastic Hessian-Free Optimization
2,013
http://arxiv.org/pdf/1301.3641v3
Title Training Neural Networks Stochastic HessianFree Optimization Summary Hessianfree HF optimization successfully used training deep autoencoders recurrent network HF us conjugate gradient algorithm construct update direction curvaturevector product computed order time gradient paper exploit property study stochastic...
[-0.02654571644961834, 0.025545494630932808, -0.00654709804803133, 0.07047495245933533, 0.027276460081338882, -0.005315929185599089, 0.02863301709294319, -0.016452424228191376, -0.022089684382081032, -0.004190881736576557, -0.018904509022831917, -0.02286473475396633, 0.022894447669386864, 0.03298916667699814, 0.0101448...
932
932
['Christophe Andrieu', 'Nando de Freitas', 'Arnaud Doucet']
1301.3833v1
We propose a novel reversible jump Markov chain Monte Carlo (MCMC) simulated annealing algorithm to optimize radial basis function (RBF) networks. This algorithm enables us to maximize the joint posterior distribution of the network parameters and the number of basis functions. It performs a global search in the joint ...
Reversible Jump MCMC Simulated Annealing for Neural Networks
2,013
http://arxiv.org/pdf/1301.3833v1
Title Reversible Jump MCMC Simulated Annealing Neural Networks Summary propose novel reversible jump Markov chain Monte Carlo MCMC simulated annealing algorithm optimize radial basis function RBF network algorithm enables u maximize joint posterior distribution network parameter number basis function performs global se...
[-0.010786466300487518, -0.009773246012628078, -0.018252236768603325, -0.04224435240030289, -0.0017850991571322083, -0.015829797834157944, -0.07046064734458923, -0.0033105129841715097, -0.03357561305165291, 0.01090709026902914, -0.04411371052265167, 0.02305375039577484, -0.00020831477013416588, 0.022349296137690544, 0....
933
933
['Misha Denil', 'Babak Shakibi', 'Laurent Dinh', "Marc'Aurelio Ranzato", 'Nando de Freitas']
1306.0543v2
We demonstrate that there is significant redundancy in the parameterization of several deep learning models. Given only a few weight values for each feature it is possible to accurately predict the remaining values. Moreover, we show that not only can the parameter values be predicted, but many of them need not be lear...
Predicting Parameters in Deep Learning
2,013
http://arxiv.org/pdf/1306.0543v2
Title Predicting Parameters Deep Learning Summary demonstrate significant redundancy parameterization several deep learning model Given weight value feature possible accurately predict remaining value Moreover show parameter value predicted many need learned train several different architecture learning small number we...
[0.0017769146943464875, 0.045442476868629456, -0.012156003154814243, -0.0026195708196610212, 0.053141359239816666, -0.021875018253922462, 0.03320778161287308, -0.004966241307556629, -0.025984233245253563, 0.04710373654961586, 0.013110204599797726, 0.004445841070264578, 0.016273245215415955, 0.06793177127838135, 0.01967...
934
934
['Guillaume Desjardins', 'Aaron Courville', 'Yoshua Bengio']
1210.5474v1
Here we propose a novel model family with the objective of learning to disentangle the factors of variation in data. Our approach is based on the spike-and-slab restricted Boltzmann machine which we generalize to include higher-order interactions among multiple latent variables. Seen from a generative perspective, the ...
Disentangling Factors of Variation via Generative Entangling
2,012
http://arxiv.org/pdf/1210.5474v1
Title Disentangling Factors Variation via Generative Entangling Summary propose novel model family objective learning disentangle factor variation data approach based spikeandslab restricted Boltzmann machine generalize include higherorder interaction among multiple latent variable Seen generative perspective multiplic...
[0.01148148626089096, 0.044371139258146286, -0.017013508826494217, 0.0487491749227047, 0.03362919017672539, 0.058040469884872437, 0.03184833005070686, 0.03132623806595802, -0.0501619428396225, 0.04874086380004883, 0.015379191376268864, -0.008634132333099842, -0.001956221181899309, 0.09703218191862106, 0.010026331059634...
935
935
['Marie Cottrell', 'Madalina Olteanu', 'Fabrice Rossi', 'Joseph Rynkiewicz', 'Nathalie Villa-Vialaneix']
1210.6511v1
Artificial neural networks are simple and efficient machine learning tools. Defined originally in the traditional setting of simple vector data, neural network models have evolved to address more and more difficulties of complex real world problems, ranging from time evolving data to sophisticated data structures such ...
Neural Networks for Complex Data
2,012
http://arxiv.org/pdf/1210.6511v1
Title Neural Networks Complex Data Summary Artificial neural network simple efficient machine learning tool Defined originally traditional setting simple vector data neural network model evolved address difficulty complex real world problem ranging time evolving data sophisticated data structure graph function paper su...
[-0.034978121519088745, 0.04601696506142616, -0.039164163172245026, -3.180909698130563e-05, -0.0012761772377416492, -0.026541996747255325, 0.03580918535590172, -0.016084803268313408, -0.034720487892627716, 0.01704436168074608, 0.03706144541501999, 0.005556359421461821, 0.03033619001507759, 0.047243595123291016, 0.01455...
936
936
['George E. Dahl', 'Navdeep Jaitly', 'Ruslan Salakhutdinov']
1406.1231v1
Although artificial neural networks have occasionally been used for Quantitative Structure-Activity/Property Relationship (QSAR/QSPR) studies in the past, the literature has of late been dominated by other machine learning techniques such as random forests. However, a variety of new neural net techniques along with suc...
Multi-task Neural Networks for QSAR Predictions
2,014
http://arxiv.org/pdf/1406.1231v1
Title Multitask Neural Networks QSAR Predictions Summary Although artificial neural network occasionally used Quantitative StructureActivityProperty Relationship QSARQSPR study past literature late dominated machine learning technique random forest However variety new neural net technique along successful application d...
[-0.004167737439274788, 0.0201212540268898, -0.03157481923699379, -0.021820642054080963, 0.009284436702728271, -0.024197613820433617, 0.03925786167383194, 0.0022811272647231817, 0.06030399724841118, 0.03369792550802231, 0.013195612467825413, 0.0020212747622281313, -0.006949395872652531, 0.05171136185526848, 0.037571411...
937
937
['Michael R. Smith', 'Tony Martinez', 'Michael Gashler']
1406.2235v1
Collaborative filtering is used to recommend items to a user without requiring a knowledge of the item itself and tends to outperform other techniques. However, collaborative filtering suffers from the cold-start problem, which occurs when an item has not yet been rated or a user has not rated any items. Incorporating ...
A Hybrid Latent Variable Neural Network Model for Item Recommendation
2,014
http://arxiv.org/pdf/1406.2235v1
Title Hybrid Latent Variable Neural Network Model Item Recommendation Summary Collaborative filtering used recommend item user without requiring knowledge item tends outperform technique However collaborative filtering suffers coldstart problem occurs item yet rated user rated item Incorporating additional information ...
[0.02611149288713932, -0.021026501432061195, 0.0049522374756634235, 0.005302788689732552, -0.01011070515960455, -0.0007123907562345266, 0.027500642463564873, 0.0023371907882392406, -0.03531720116734505, -0.013227098621428013, -0.01252741925418377, 0.003520858706906438, -0.00421498715877533, 0.0906575471162796, -0.00265...
938
938
['Tapani Raiko', 'Mathias Berglund', 'Guillaume Alain', 'Laurent Dinh']
1406.2989v3
Stochastic binary hidden units in a multi-layer perceptron (MLP) network give at least three potential benefits when compared to deterministic MLP networks. (1) They allow to learn one-to-many type of mappings. (2) They can be used in structured prediction problems, where modeling the internal structure of the output i...
Techniques for Learning Binary Stochastic Feedforward Neural Networks
2,014
http://arxiv.org/pdf/1406.2989v3
Title Techniques Learning Binary Stochastic Feedforward Neural Networks Summary Stochastic binary hidden unit multilayer perceptron MLP network give least three potential benefit compared deterministic MLP network 1 allow learn onetomany type mapping 2 used structured prediction problem modeling internal structure outp...
[0.001215980271808803, 0.04254048317670822, 0.00035428415867500007, 0.008588207885622978, -0.002210453851148486, -0.028770996257662773, 0.05483630299568176, 0.014176048338413239, 0.01496941689401865, 0.004779358860105276, -0.0037284898571670055, -0.003071877406910062, -0.0034792001824826, 0.07418079674243927, 0.0613628...
939
939
['Philip de Chazal', 'Jonathan Tapson', 'André van Schaik']
1406.3100v1
We present an alternative to the pseudo-inverse method for determining the hidden to output weight values for Extreme Learning Machines performing classification tasks. The method is based on linear discriminant analysis and provides Bayes optimal single point estimates for the weight values.
Learning ELM network weights using linear discriminant analysis
2,014
http://arxiv.org/pdf/1406.3100v1
Title Learning ELM network weight using linear discriminant analysis Summary present alternative pseudoinverse method determining hidden output weight value Extreme Learning Machines performing classification task method based linear discriminant analysis provides Bayes optimal single point estimate weight value Author...
[-0.047780271619558334, -0.05507497489452362, -0.04572468250989914, 0.022593537345528603, 0.03775206580758095, 0.000574636273086071, 0.008601496927440166, -0.0022422426845878363, 0.01774703525006771, 0.04359229654073715, 0.0014751049457117915, 0.05325311794877052, -0.0011684485943987966, 0.05012005195021629, 0.02129371...
940
940
['Kyunghyun Cho', 'Yoshua Bengio']
1406.7362v1
Many state-of-the-art results obtained with deep networks are achieved with the largest models that could be trained, and if more computation power was available, we might be able to exploit much larger datasets in order to improve generalization ability. Whereas in learning algorithms such as decision trees the ratio ...
Exponentially Increasing the Capacity-to-Computation Ratio for Conditional Computation in Deep Learning
2,014
http://arxiv.org/pdf/1406.7362v1
Title Exponentially Increasing CapacitytoComputation Ratio Conditional Computation Deep Learning Summary Many stateoftheart result obtained deep network achieved largest model could trained computation power available might able exploit much larger datasets order improve generalization ability Whereas learning algorith...
[0.010267950594425201, 0.051133278757333755, -0.03516475856304169, 0.03024945966899395, 0.03197839483618736, 0.0034251566976308823, 0.007188150193542242, -0.004338458180427551, -0.05059635266661644, 0.00571111124008894, -0.004952612798660994, -0.012782791629433632, -0.024504205211997032, 0.03483777865767479, -0.0040868...
941
941
['Taichi Kiwaki']
1505.02462v3
We present a layered Boltzmann machine (BM) that can better exploit the advantages of a distributed representation. It is widely believed that deep BMs (DBMs) have far greater representational power than its shallow counterpart, restricted Boltzmann machines (RBMs). However, this expectation on the supremacy of DBMs ov...
Soft-Deep Boltzmann Machines
2,015
http://arxiv.org/pdf/1505.02462v3
Title SoftDeep Boltzmann Machines Summary present layered Boltzmann machine BM better exploit advantage distributed representation widely believed deep BMs DBMs far greater representational power shallow counterpart restricted Boltzmann machine RBMs However expectation supremacy DBMs RBMs ever validated theoretical fas...
[-0.02861926145851612, -0.03383118659257889, -0.03731789067387581, 0.018691247329115868, -0.02108532004058361, 4.4967338908463717e-05, 0.09842293709516525, -0.01948707178235054, -0.058954812586307526, 0.01275632530450821, -0.029230007901787758, 0.003911531995981932, 0.00564573984593153, 0.05199882388114929, 0.032638389...
942
942
['Yaroslav Ganin', 'Evgeniya Ustinova', 'Hana Ajakan', 'Pascal Germain', 'Hugo Larochelle', 'François Laviolette', 'Mario Marchand', 'Victor Lempitsky']
1505.07818v4
We introduce a new representation learning approach for domain adaptation, in which data at training and test time come from similar but different distributions. Our approach is directly inspired by the theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predictions must be made b...
Domain-Adversarial Training of Neural Networks
2,015
http://arxiv.org/pdf/1505.07818v4
Title DomainAdversarial Training Neural Networks Summary introduce new representation learning approach domain adaptation data training test time come similar different distribution approach directly inspired theory domain adaptation suggesting effective domain transfer achieved prediction must made based feature canno...
[0.019201599061489105, 0.034077271819114685, -0.026657868176698685, 0.03172437846660614, 0.0033702924847602844, -0.0048766229301691055, 0.042705897241830826, -0.012422313913702965, -0.018111437559127808, -0.048603322356939316, -0.05782104283571243, -0.010119378566741943, -0.02643941529095173, 0.04546100273728371, -0.00...
943
943
['David Balduzzi']
1604.01952v1
Methods from convex optimization are widely used as building blocks for deep learning algorithms. However, the reasons for their empirical success are unclear, since modern convolutional networks (convnets), incorporating rectifier units and max-pooling, are neither smooth nor convex. Standard guarantees therefore do n...
Deep Online Convex Optimization with Gated Games
2,016
http://arxiv.org/pdf/1604.01952v1
Title Deep Online Convex Optimization Gated Games Summary Methods convex optimization widely used building block deep learning algorithm However reason empirical success unclear since modern convolutional network convnets incorporating rectifier unit maxpooling neither smooth convex Standard guarantee therefore apply p...
[-0.0025042917113751173, 0.028802013024687767, -0.022399017587304115, 0.042485352605581284, -0.027306577190756798, -0.037745244801044464, 0.010233091190457344, -0.01977705955505371, -0.05484988912940025, 0.023186886683106422, 0.008752051740884781, -0.029419537633657455, -0.0033265207894146442, 0.1092197373509407, 0.025...
944
944
['Artit Wangperawong', 'Cyrille Brun', 'Olav Laudy', 'Rujikorn Pavasuthipaisit']
1604.05377v1
Customer temporal behavioral data was represented as images in order to perform churn prediction by leveraging deep learning architectures prominent in image classification. Supervised learning was performed on labeled data of over 6 million customers using deep convolutional neural networks, which achieved an AUC of 0...
Churn analysis using deep convolutional neural networks and autoencoders
2,016
http://arxiv.org/pdf/1604.05377v1
Title Churn analysis using deep convolutional neural network autoencoders Summary Customer temporal behavioral data represented image order perform churn prediction leveraging deep learning architecture prominent image classification Supervised learning performed labeled data 6 million customer using deep convolutional...
[0.017682963982224464, 0.06825803965330124, -0.04299362748861313, -0.028144652023911476, 0.022813091054558754, 0.02680245414376259, 0.0631818100810051, -0.009304946288466454, -0.04799279198050499, 0.01680714637041092, 0.015827804803848267, 0.015125793404877186, -0.004157013259828091, 0.09472476691007614, -0.03587032854...
945
945
['Chee Chun Gan', 'Gerard Learmonth']
1604.06730v1
ICU mortality scoring systems attempt to predict patient mortality using predictive models with various clinical predictors. Examples of such systems are APACHE, SAPS and MPM. However, most such scoring systems do not actively look for and include interaction terms, despite physicians intuitively taking such interactio...
Developing an ICU scoring system with interaction terms using a genetic algorithm
2,016
http://arxiv.org/pdf/1604.06730v1
Title Developing ICU scoring system interaction term using genetic algorithm Summary ICU mortality scoring system attempt predict patient mortality using predictive model various clinical predictor Examples system APACHE SAPS MPM However scoring system actively look include interaction term despite physician intuitivel...
[0.022869685664772987, 0.024017179384827614, -0.013866687193512917, -0.053549669682979584, 0.0076678358018398285, 0.004625585861504078, 0.04362069442868233, 0.029417937621474266, -0.006000105757266283, 0.02139606885612011, 0.04166088625788689, -0.030953554436564445, 0.06068505719304085, 0.11424346268177032, -0.02872847...
946
946
['Henry Z. Lo', 'Kevin Amaral', 'Wei Ding']
1604.07796v1
One of the difficulties of training deep neural networks is caused by improper scaling between layers. Scaling issues introduce exploding / gradient problems, and have typically been addressed by careful scale-preserving initialization. We investigate the value of preserving scale, or isometry, beyond the initial weigh...
Scale Normalization
2,016
http://arxiv.org/pdf/1604.07796v1
Title Scale Normalization Summary One difficulty training deep neural network caused improper scaling layer Scaling issue introduce exploding gradient problem typically addressed careful scalepreserving initialization investigate value preserving scale isometry beyond initial weight propose two method maintaing isometr...
[-0.05608515068888664, 0.02231414243578911, -0.014134297147393227, -0.01099756546318531, 0.026236237958073616, 0.0034751086495816708, 0.0441448874771595, 0.008070431649684906, -0.0455341599881649, 0.062275830656290054, -0.015970904380083084, -0.036701396107673645, -0.0011890233727172017, 0.07825721800327301, -0.0166249...
947
947
['Ludovic Arnold', 'Yann Ollivier']
1212.1524v2
When using deep, multi-layered architectures to build generative models of data, it is difficult to train all layers at once. We propose a layer-wise training procedure admitting a performance guarantee compared to the global optimum. It is based on an optimistic proxy of future performance, the best latent marginal. W...
Layer-wise learning of deep generative models
2,012
http://arxiv.org/pdf/1212.1524v2
Title Layerwise learning deep generative model Summary using deep multilayered architecture build generative model data difficult train layer propose layerwise training procedure admitting performance guarantee compared global optimum based optimistic proxy future performance best latent marginal interpret autoencoders...
[-0.0313352532684803, 0.06620785593986511, -0.0353248156607151, 0.013033506460487843, 0.02999725565314293, -0.01258664671331644, 0.06161899492144585, -0.03296348825097084, -0.048237282782793045, 0.03418130800127983, 0.019652310758829117, 0.03305657207965851, -0.01949579082429409, 0.0628538578748703, 0.05576291680335998...
948
948
['Miguel Á. Carreira-Perpiñán', 'Weiran Wang']
1212.5921v1
In science and engineering, intelligent processing of complex signals such as images, sound or language is often performed by a parameterized hierarchy of nonlinear processing layers, sometimes biologically inspired. Hierarchical systems (or, more generally, nested systems) offer a way to generate complex mappings usin...
Distributed optimization of deeply nested systems
2,012
http://arxiv.org/pdf/1212.5921v1
Title Distributed optimization deeply nested system Summary science engineering intelligent processing complex signal image sound language often performed parameterized hierarchy nonlinear processing layer sometimes biologically inspired Hierarchical system generally nested system offer way generate complex mapping usi...
[0.0008468810701742768, 0.06830428540706635, -0.03854633495211601, 0.028634129092097282, 0.0020932983607053757, -0.010270696133375168, 0.06676343083381653, -0.0018370755715295672, -0.0471697635948658, -0.017862211912870407, -0.056982871145009995, 0.009396135807037354, 0.024866638705134392, 0.025313515216112137, -0.0032...
949
949
['Xiaozhao Zhao', 'Yuexian Hou', 'Qian Yu', 'Dawei Song', 'Wenjie Li']
1302.3931v7
Typical dimensionality reduction methods focus on directly reducing the number of random variables while retaining maximal variations in the data. In this paper, we consider the dimensionality reduction in parameter spaces of binary multivariate distributions. We propose a general Confident-Information-First (CIF) prin...
Understanding Boltzmann Machine and Deep Learning via A Confident Information First Principle
2,013
http://arxiv.org/pdf/1302.3931v7
Title Understanding Boltzmann Machine Deep Learning via Confident Information First Principle Summary Typical dimensionality reduction method focus directly reducing number random variable retaining maximal variation data paper consider dimensionality reduction parameter space binary multivariate distribution propose g...
[-0.025464095175266266, -0.013774629682302475, -0.042159441858530045, 0.027779314666986465, -0.016197023913264275, -0.026659326627850533, 0.06105431169271469, -0.004128093831241131, -0.06401896476745605, 0.057078924030065536, 0.012049119919538498, -0.004306909162551165, 0.008740131743252277, 0.07568143308162689, 0.0347...
950
950
['Vittorio Latorre', 'David Yang Gao']
1302.4141v1
Radial Basis Functions Neural Networks (RBFNNs) are tools widely used in regression problems. One of their principal drawbacks is that the formulation corresponding to the training with the supervision of both the centers and the weights is a highly non-convex optimization problem, which leads to some fundamentally dif...
Canonical dual solutions to nonconvex radial basis neural network optimization problem
2,013
http://arxiv.org/pdf/1302.4141v1
Title Canonical dual solution nonconvex radial basis neural network optimization problem Summary Radial Basis Functions Neural Networks RBFNNs tool widely used regression problem One principal drawback formulation corresponding training supervision center weight highly nonconvex optimization problem lead fundamentally ...
[-0.014607925899326801, 0.024343496188521385, -0.02844329923391342, 0.040786631405353546, 0.01724657043814659, -0.012980449013411999, -0.0007429387187585235, -0.0013303601881489158, -0.040544718503952026, 0.007509095128625631, -0.02854190394282341, -0.01185290701687336, -0.004880169872194529, 0.009530519135296345, 0.01...
951
951
['Guido Montúfar', 'Razvan Pascanu', 'Kyunghyun Cho', 'Yoshua Bengio']
1402.1869v2
We study the complexity of functions computable by deep feedforward neural networks with piecewise linear activations in terms of the symmetries and the number of linear regions that they have. Deep networks are able to sequentially map portions of each layer's input-space to the same output. In this way, deep models c...
On the Number of Linear Regions of Deep Neural Networks
2,014
http://arxiv.org/pdf/1402.1869v2
Title Number Linear Regions Deep Neural Networks Summary study complexity function computable deep feedforward neural network piecewise linear activation term symmetry number linear region Deep network able sequentially map portion layer inputspace output way deep model compute function react equally complicated patter...
[-0.015910517424345016, 0.026597000658512115, -0.04318666458129883, 0.04936468228697777, -0.03697515279054642, -0.014711592346429825, 0.06089446693658829, -0.03369078412652016, -0.035225607454776764, 0.00869825016707182, 0.015742894262075424, -0.003113511949777603, 0.009672436863183975, 0.09304176270961761, 0.016840938...
952
952
['Guido Montufar', 'Nihat Ay', 'Keyan Ghazi-Zahedi']
1402.3346v3
Conditional restricted Boltzmann machines are undirected stochastic neural networks with a layer of input and output units connected bipartitely to a layer of hidden units. These networks define models of conditional probability distributions on the states of the output units given the states of the input units, parame...
Geometry and Expressive Power of Conditional Restricted Boltzmann Machines
2,014
http://arxiv.org/pdf/1402.3346v3
Title Geometry Expressive Power Conditional Restricted Boltzmann Machines Summary Conditional restricted Boltzmann machine undirected stochastic neural network layer input output unit connected bipartitely layer hidden unit network define model conditional probability distribution state output unit given state input un...
[-0.03153286874294281, -0.010947375558316708, -0.03120209462940693, -0.011755645275115967, -0.03535754978656769, -0.02984841912984848, 0.02683355286717415, -0.03830824792385101, -0.020294515416026115, -0.00038627846515737474, 0.021498078480362892, -0.0029827968683093786, 0.0050195432268083096, 0.047754984349012375, 0.0...
953
953
['Yingbo Zhou', 'Devansh Arpit', 'Ifeoma Nwogu', 'Venu Govindaraju']
1405.1380v4
Traditionally, when generative models of data are developed via deep architectures, greedy layer-wise pre-training is employed. In a well-trained model, the lower layer of the architecture models the data distribution conditional upon the hidden variables, while the higher layers model the hidden distribution prior. Bu...
Is Joint Training Better for Deep Auto-Encoders?
2,014
http://arxiv.org/pdf/1405.1380v4
Title Joint Training Better Deep AutoEncoders Summary Traditionally generative model data developed via deep architecture greedy layerwise pretraining employed welltrained model lower layer architecture model data distribution conditional upon hidden variable higher layer model hidden distribution prior due greedy sche...
[-0.012143359519541264, 0.09947140514850616, -0.027615344151854515, 0.04559614136815071, 0.035100433975458145, -0.005233585834503174, 0.06767681241035461, -0.0028550676070153713, -0.051131196320056915, 0.008130588568747044, -0.045346204191446304, 0.01610741578042507, -0.000984018319286406, 0.06274141371250153, 0.017853...
954
954
['Bharath Ramsundar', 'Steven Kearnes', 'Patrick Riley', 'Dale Webster', 'David Konerding', 'Vijay Pande']
1502.02072v1
Massively multitask neural architectures provide a learning framework for drug discovery that synthesizes information from many distinct biological sources. To train these architectures at scale, we gather large amounts of data from public sources to create a dataset of nearly 40 million measurements across more than 2...
Massively Multitask Networks for Drug Discovery
2,015
http://arxiv.org/pdf/1502.02072v1
Title Massively Multitask Networks Drug Discovery Summary Massively multitask neural architecture provide learning framework drug discovery synthesizes information many distinct biological source train architecture scale gather large amount data public source create dataset nearly 40 million measurement across 200 biol...
[-0.0063495696522295475, 0.04975642263889313, -0.04969564825296402, -0.02927136979997158, 0.023149767890572548, 0.01348835974931717, 0.05887255072593689, -0.004094231873750687, 0.03089132159948349, 0.031276870518922806, -0.04117029160261154, 0.006186211481690407, -0.054234955459833145, 0.09641271084547043, 0.0416954271...
955
955
['Junyoung Chung', 'Caglar Gulcehre', 'Kyunghyun Cho', 'Yoshua Bengio']
1502.02367v4
In this work, we propose a novel recurrent neural network (RNN) architecture. The proposed RNN, gated-feedback RNN (GF-RNN), extends the existing approach of stacking multiple recurrent layers by allowing and controlling signals flowing from upper recurrent layers to lower layers using a global gating unit for each pai...
Gated Feedback Recurrent Neural Networks
2,015
http://arxiv.org/pdf/1502.02367v4
Title Gated Feedback Recurrent Neural Networks Summary work propose novel recurrent neural network RNN architecture proposed RNN gatedfeedback RNN GFRNN extends existing approach stacking multiple recurrent layer allowing controlling signal flowing upper recurrent layer lower layer using global gating unit pair layer r...
[0.010986789129674435, 0.02426208183169365, 0.015270414762198925, 0.050046589225530624, -0.002130315639078617, -0.008288975805044174, 0.014075572602450848, -0.009453745558857918, -0.006841693539172411, -0.013927253894507885, -0.034967657178640366, -0.0526738166809082, 0.049283236265182495, 0.05515960603952408, 0.013958...
956
956
['Suyog Gupta', 'Ankur Agrawal', 'Kailash Gopalakrishnan', 'Pritish Narayanan']
1502.02551v1
Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a cruc...
Deep Learning with Limited Numerical Precision
2,015
http://arxiv.org/pdf/1502.02551v1
Title Deep Learning Limited Numerical Precision Summary Training largescale deep neural network often constrained available computational resource study effect limited precision data representation computation neural network training Within context lowprecision fixedpoint computation observe rounding scheme play crucia...
[-0.049684226512908936, 0.01529654860496521, -0.028503447771072388, 0.026530034840106964, 0.023879896849393845, -0.025578662753105164, 0.06392213702201843, -0.020200366154313087, -0.06457959115505219, -0.0004043357621412724, -0.03333777189254761, -0.05165514349937439, 0.03641494736075401, 0.03819923475384712, 0.0105865...
957
957
['Mathieu Germain', 'Karol Gregor', 'Iain Murray', 'Hugo Larochelle']
1502.03509v2
There has been a lot of recent interest in designing neural network models to estimate a distribution from a set of examples. We introduce a simple modification for autoencoder neural networks that yields powerful generative models. Our method masks the autoencoder's parameters to respect autoregressive constraints: ea...
MADE: Masked Autoencoder for Distribution Estimation
2,015
http://arxiv.org/pdf/1502.03509v2
Title MADE Masked Autoencoder Distribution Estimation Summary lot recent interest designing neural network model estimate distribution set example introduce simple modification autoencoder neural network yield powerful generative model method mask autoencoders parameter respect autoregressive constraint input reconstru...
[-0.022834602743387222, 0.09871013462543488, -0.05054556950926781, 0.019892383366823196, 0.013580773025751114, -0.03144886717200279, 0.051417261362075806, -0.011388998478651047, -0.0722624883055687, 0.0027885737363249063, 0.03075500577688217, 0.018086010590195656, 0.035774264484643936, 0.09353811293840408, 0.0200691521...
958
958
['Sanjeev Arora', 'Rong Ge', 'Tengyu Ma', 'Ankur Moitra']
1503.00778v1
Sparse coding is a basic task in many fields including signal processing, neuroscience and machine learning where the goal is to learn a basis that enables a sparse representation of a given set of data, if one exists. Its standard formulation is as a non-convex optimization problem which is solved in practice by heuri...
Simple, Efficient, and Neural Algorithms for Sparse Coding
2,015
http://arxiv.org/pdf/1503.00778v1
Title Simple Efficient Neural Algorithms Sparse Coding Summary Sparse coding basic task many field including signal processing neuroscience machine learning goal learn basis enables sparse representation given set data one exists standard formulation nonconvex optimization problem solved practice heuristic based altern...
[-0.0086577283218503, 0.052953846752643585, 0.018169106915593147, 0.019068920984864235, 0.0068247378803789616, 0.01782236434519291, -0.007205440662801266, 0.0614737905561924, -0.05025531351566315, -0.006142206024378538, 0.0034828209318220615, -0.012042325921356678, 0.0408938005566597, 0.034481193870306015, 0.0286012236...
959
959
['Thomas Unterthiner', 'Andreas Mayr', 'Günter Klambauer', 'Sepp Hochreiter']
1503.01445v1
Everyday we are exposed to various chemicals via food additives, cleaning and cosmetic products and medicines -- and some of them might be toxic. However testing the toxicity of all existing compounds by biological experiments is neither financially nor logistically feasible. Therefore the government agencies NIH, EPA ...
Toxicity Prediction using Deep Learning
2,015
http://arxiv.org/pdf/1503.01445v1
Title Toxicity Prediction using Deep Learning Summary Everyday exposed various chemical via food additive cleaning cosmetic product medicine might toxic However testing toxicity existing compound biological experiment neither financially logistically feasible Therefore government agency NIH EPA FDA launched Tox21 Data ...
[-0.0007444783113896847, 0.04650363326072693, -0.016212070360779762, -0.03169908747076988, 0.030972357839345932, 0.009790626354515553, 0.043538421392440796, 0.03996269404888153, 0.04004068672657013, 0.024691862985491753, 0.00705236429348588, 0.04221075400710106, -0.03909539431333542, 0.08947142958641052, 0.036346625536...
960
960
['Prateek Jain', 'Vivek Kulkarni', 'Abhradeep Thakurta', 'Oliver Williams']
1503.02031v1
Training deep belief networks (DBNs) requires optimizing a non-convex function with an extremely large number of parameters. Naturally, existing gradient descent (GD) based methods are prone to arbitrarily poor local minima. In this paper, we rigorously show that such local minima can be avoided (upto an approximation ...
To Drop or Not to Drop: Robustness, Consistency and Differential Privacy Properties of Dropout
2,015
http://arxiv.org/pdf/1503.02031v1
Title Drop Drop Robustness Consistency Differential Privacy Properties Dropout Summary Training deep belief network DBNs requires optimizing nonconvex function extremely large number parameter Naturally existing gradient descent GD based method prone arbitrarily poor local minimum paper rigorously show local minimum av...
[-0.017805472016334534, 0.09141001850366592, -0.030749650672078133, 0.042688459157943726, -0.03443252295255661, -0.04982313513755798, 0.06820384413003922, -0.051053207367658615, 0.018832944333553314, 0.03526526317000389, -0.01326978113502264, 0.024896224960684776, -0.003951945807784796, 0.08358075469732285, -0.01370614...
961
961
['Geoffrey Hinton', 'Oriol Vinyals', 'Jeff Dean']
1503.02531v1
A very simple way to improve the performance of almost any machine learning algorithm is to train many different models on the same data and then to average their predictions. Unfortunately, making predictions using a whole ensemble of models is cumbersome and may be too computationally expensive to allow deployment to...
Distilling the Knowledge in a Neural Network
2,015
http://arxiv.org/pdf/1503.02531v1
Title Distilling Knowledge Neural Network Summary simple way improve performance almost machine learning algorithm train many different model data average prediction Unfortunately making prediction using whole ensemble model cumbersome may computationally expensive allow deployment large number user especially individu...
[-0.028798166662454605, 0.05130145326256752, 0.004625328816473484, 0.04425036162137985, -0.011224078945815563, -0.01668057218194008, 0.016323305666446686, 0.04563092812895775, -0.05150187760591507, 0.007483021821826696, -0.041148751974105835, 0.016859106719493866, 0.029475035145878792, 0.015684306621551514, 0.002764841...
962
962
['Joan Bruna', 'Soumith Chintala', 'Yann LeCun', 'Serkan Piantino', 'Arthur Szlam', 'Mark Tygert']
1503.03438v3
A complex-valued convolutional network (convnet) implements the repeated application of the following composition of three operations, recursively applying the composition to an input vector of nonnegative real numbers: (1) convolution with complex-valued vectors followed by (2) taking the absolute value of every entry...
A mathematical motivation for complex-valued convolutional networks
2,015
http://arxiv.org/pdf/1503.03438v3
Title mathematical motivation complexvalued convolutional network Summary complexvalued convolutional network convnet implement repeated application following composition three operation recursively applying composition input vector nonnegative real number 1 convolution complexvalued vector followed 2 taking absolute v...
[-0.044492967426776886, 0.03256741166114807, 0.00335773010738194, -0.015093979425728321, -0.026445599272847176, 0.003916062880307436, 0.047120507806539536, -0.028478853404521942, -0.07745401561260223, 0.018042949959635735, 0.021035758778452873, -0.02250479720532894, 0.00202585244551301, 0.029762981459498405, 0.03109667...
963
963
['James Martens', 'Roger Grosse']
1503.05671v6
We propose an efficient method for approximating natural gradient descent in neural networks which we call Kronecker-Factored Approximate Curvature (K-FAC). K-FAC is based on an efficiently invertible approximation of a neural network's Fisher information matrix which is neither diagonal nor low-rank, and in some cases...
Optimizing Neural Networks with Kronecker-factored Approximate Curvature
2,015
http://arxiv.org/pdf/1503.05671v6
Title Optimizing Neural Networks Kroneckerfactored Approximate Curvature Summary propose efficient method approximating natural gradient descent neural network call KroneckerFactored Approximate Curvature KFAC KFAC based efficiently invertible approximation neural network Fisher information matrix neither diagonal lowr...
[-0.012532608583569527, 0.014184918254613876, 0.0036409306339919567, 0.058472055941820145, 0.004249291494488716, -0.02122035063803196, -0.00973591860383749, 0.009059455245733261, -0.028992824256420135, 0.0506318099796772, -0.018538257107138634, 0.011797339655458927, 0.011298962868750095, -0.01199228037148714, 0.0418176...
964
964
['Xiao-Lei Zhang']
1503.06452v1
Recently, multilayer bootstrap network (MBN) has demonstrated promising performance in unsupervised dimensionality reduction. It can learn compact representations in standard data sets, i.e. MNIST and RCV1. However, as a bootstrap method, the prediction complexity of MBN is high. In this paper, we propose an unsupervis...
Unsupervised model compression for multilayer bootstrap networks
2,015
http://arxiv.org/pdf/1503.06452v1
Title Unsupervised model compression multilayer bootstrap network Summary Recently multilayer bootstrap network MBN demonstrated promising performance unsupervised dimensionality reduction learn compact representation standard data set ie MNIST RCV1 However bootstrap method prediction complexity MBN high paper propose ...
[-0.05592220276594162, 0.06550544500350952, -0.026258407160639763, -0.01240052655339241, -0.017331624403595924, -0.009575430303812027, 0.03874025121331215, 0.04392075166106224, -0.04054724797606468, 0.0416618250310421, -0.030330320820212364, 0.019403638318181038, 0.02352558635175228, 0.039835456758737564, 0.02974588796...
965
965
['Leen De Baets', 'Joeri Ruyssinck', 'Thomas Peiffer', 'Johan Decruyenaere', 'Filip De Turck', 'Femke Ongenae', 'Tom Dhaene']
1612.00962v1
The presence of bacteria or fungi in the bloodstream of patients is abnormal and can lead to life-threatening conditions. A computational model based on a bidirectional long short-term memory artificial neural network, is explored to assist doctors in the intensive care unit to predict whether examination of blood cult...
Positive blood culture detection in time series data using a BiLSTM network
2,016
http://arxiv.org/pdf/1612.00962v1
Title Positive blood culture detection time series data using BiLSTM network Summary presence bacteria fungi bloodstream patient abnormal lead lifethreatening condition computational model based bidirectional long shortterm memory artificial neural network explored assist doctor intensive care unit predict whether exam...
[-0.012951232492923737, -0.008543558418750763, -0.031216444447636604, -0.031555403023958206, -0.0752585157752037, -0.02529570832848549, 0.03065352700650692, 0.00014822077355347574, 0.026094481348991394, 0.031777892261743546, 0.0319533497095108, -0.058320336043834686, -0.004646113608032465, 0.12846587598323822, -0.00112...
966
966
['Ramon Oliveira', 'Pedro Tabacof', 'Eduardo Valle']
1612.01251v2
We evaluate the uncertainty quality in neural networks using anomaly detection. We extract uncertainty measures (e.g. entropy) from the predictions of candidate models, use those measures as features for an anomaly detector, and gauge how well the detector differentiates known from unknown classes. We assign higher unc...
Known Unknowns: Uncertainty Quality in Bayesian Neural Networks
2,016
http://arxiv.org/pdf/1612.01251v2
Title Known Unknowns Uncertainty Quality Bayesian Neural Networks Summary evaluate uncertainty quality neural network using anomaly detection extract uncertainty measure eg entropy prediction candidate model use measure feature anomaly detector gauge well detector differentiates known unknown class assign higher uncert...
[0.0036830634344369173, 0.08199070394039154, -0.010826492682099342, 0.004590068478137255, -0.00260705198161304, -0.02226206474006176, 0.019099324941635132, 0.007932019419968128, -0.06679008156061172, 0.024930398911237717, 0.07076667994260788, -0.02156977355480194, 0.012950809672474861, 0.044534120708703995, 0.002035190...
967
967
['Tan Nguyen', 'Wanjia Liu', 'Ethan Perez', 'Richard G. Baraniuk', 'Ankit B. Patel']
1612.01942v1
Semi-supervised learning algorithms reduce the high cost of acquiring labeled training data by using both labeled and unlabeled data during learning. Deep Convolutional Networks (DCNs) have achieved great success in supervised tasks and as such have been widely employed in the semi-supervised learning. In this paper we...
Semi-Supervised Learning with the Deep Rendering Mixture Model
2,016
http://arxiv.org/pdf/1612.01942v1
Title SemiSupervised Learning Deep Rendering Mixture Model Summary Semisupervised learning algorithm reduce high cost acquiring labeled training data using labeled unlabeled data learning Deep Convolutional Networks DCNs achieved great success supervised task widely employed semisupervised learning paper leverage recen...
[-0.008170673623681068, 0.025109563022851944, 0.037925224751234055, 0.043754056096076965, 0.02285914681851864, -0.0004681906139012426, 0.03678205609321594, -0.0021924348548054695, -0.04152170196175575, 0.028147898614406586, -0.00946744903922081, 0.022286899387836456, -0.01147459726780653, 0.01926068589091301, 0.0346119...
968
968
['Yuansi Chen', 'Cengiz Pehlevan', 'Dmitri B. Chklovskii']
1612.03480v1
Recently, a novel family of biologically plausible online algorithms for reducing the dimensionality of streaming data has been derived from the similarity matching principle. In these algorithms, the number of output dimensions can be determined adaptively by thresholding the singular values of the input data matrix. ...
Self-calibrating Neural Networks for Dimensionality Reduction
2,016
http://arxiv.org/pdf/1612.03480v1
Title Selfcalibrating Neural Networks Dimensionality Reduction Summary Recently novel family biologically plausible online algorithm reducing dimensionality streaming data derived similarity matching principle algorithm number output dimension determined adaptively thresholding singular value input data matrix However ...
[-0.05297711491584778, 0.00559360021725297, -0.033285997807979584, 0.0115312235429883, 0.02226196974515915, -0.005606521386653185, 0.06173083558678627, 0.007216775789856911, 0.024872606620192528, 0.041052237153053284, -0.041381120681762695, 0.03039839118719101, 0.007977558299899101, 0.08106248825788498, -0.002376965247...
969
969
['Li Jing', 'Yichen Shen', 'Tena Dubček', 'John Peurifoy', 'Scott Skirlo', 'Yann LeCun', 'Max Tegmark', 'Marin Soljačić']
1612.05231v3
Using unitary (instead of general) matrices in artificial neural networks (ANNs) is a promising way to solve the gradient explosion/vanishing problem, as well as to enable ANNs to learn long-term correlations in the data. This approach appears particularly promising for Recurrent Neural Networks (RNNs). In this work, w...
Tunable Efficient Unitary Neural Networks (EUNN) and their application to RNNs
2,016
http://arxiv.org/pdf/1612.05231v3
Title Tunable Efficient Unitary Neural Networks EUNN application RNNs Summary Using unitary instead general matrix artificial neural network ANNs promising way solve gradient explosionvanishing problem well enable ANNs learn longterm correlation data approach appears particularly promising Recurrent Neural Networks RNN...
[-0.009237864054739475, 0.06816868484020233, 0.0008616929408162832, 0.03955705463886261, 0.0034008780494332314, 0.005337234120815992, 0.04632338881492615, -0.02107185311615467, -0.034575048834085464, -0.008716028183698654, -0.06766808032989502, -0.047045931220054626, 0.010109322145581245, -0.012877666391432285, 0.00846...
970
970
['Alex Graves']
1211.3711v1
Many machine learning tasks can be expressed as the transformation---or \emph{transduction}---of input sequences into output sequences: speech recognition, machine translation, protein secondary structure prediction and text-to-speech to name but a few. One of the key challenges in sequence transduction is learning to ...
Sequence Transduction with Recurrent Neural Networks
2,012
http://arxiv.org/pdf/1211.3711v1
Title Sequence Transduction Recurrent Neural Networks Summary Many machine learning task expressed transformationor emphtransductionof input sequence output sequence speech recognition machine translation protein secondary structure prediction texttospeech name One key challenge sequence transduction learning represent...
[0.03449882194399834, 0.03490511700510979, 0.008206603117287159, 0.022779539227485657, -0.01845366321504116, -0.024564433842897415, -0.00438099168241024, 0.0035732174292206764, -0.05922279506921768, -0.02199256420135498, -0.011115418747067451, -0.02383965626358986, 0.05342552065849304, 0.030779361724853516, 0.034506436...
971
971
['Justin Bayer', 'Christian Osendorfer', 'Daniela Korhammer', 'Nutan Chen', 'Sebastian Urban', 'Patrick van der Smagt']
1311.0701v7
Recurrent Neural Networks (RNNs) are rich models for the processing of sequential data. Recent work on advancing the state of the art has been focused on the optimization or modelling of RNNs, mostly motivated by adressing the problems of the vanishing and exploding gradients. The control of overfitting has seen consid...
On Fast Dropout and its Applicability to Recurrent Networks
2,013
http://arxiv.org/pdf/1311.0701v7
Title Fast Dropout Applicability Recurrent Networks Summary Recurrent Neural Networks RNNs rich model processing sequential data Recent work advancing state art focused optimization modelling RNNs mostly motivated adressing problem vanishing exploding gradient control overfitting seen considerably le attention paper co...
[-0.024142397567629814, 0.002736502792686224, -0.014516575261950493, 0.03279123082756996, -0.006244538351893425, -0.026343349367380142, 0.04186433181166649, -0.0030760816298425198, -0.03644561395049095, 0.021019943058490753, -0.040526412427425385, 0.036918289959430695, -0.00025931018171831965, 0.07496020942926407, 0.02...
972
972
['Caglar Gulcehre', 'Kyunghyun Cho', 'Razvan Pascanu', 'Yoshua Bengio']
1311.1780v7
In this paper we propose and investigate a novel nonlinear unit, called $L_p$ unit, for deep neural networks. The proposed $L_p$ unit receives signals from several projections of a subset of units in the layer below and computes a normalized $L_p$ norm. We notice two interesting interpretations of the $L_p$ unit. First...
Learned-Norm Pooling for Deep Feedforward and Recurrent Neural Networks
2,013
http://arxiv.org/pdf/1311.1780v7
Title LearnedNorm Pooling Deep Feedforward Recurrent Neural Networks Summary paper propose investigate novel nonlinear unit called Lp unit deep neural network proposed Lp unit receives signal several projection subset unit layer computes normalized Lp norm notice two interesting interpretation Lp unit First proposed un...
[-0.0314551442861557, -0.0031415289267897606, 0.009956634603440762, 0.08124497532844543, 0.003736332291737199, -0.010780435055494308, 0.07215265184640884, -0.033342182636260986, -0.023948101326823235, 0.02725079655647278, -0.01886402815580368, -0.05439789965748787, -0.008117456920444965, 0.047099582850933075, 0.0138787...
973
973
['Michael S. Gashler', 'Michael R. Smith', 'Richard Morris', 'Tony Martinez']
1312.5394v1
Many data mining and data analysis techniques operate on dense matrices or complete tables of data. Real-world data sets, however, often contain unknown values. Even many classification algorithms that are designed to operate with missing values still exhibit deteriorated accuracy. One approach to handling missing valu...
Missing Value Imputation With Unsupervised Backpropagation
2,013
http://arxiv.org/pdf/1312.5394v1
Title Missing Value Imputation Unsupervised Backpropagation Summary Many data mining data analysis technique operate dense matrix complete table data Realworld data set however often contain unknown value Even many classification algorithm designed operate missing value still exhibit deteriorated accuracy One approach ...
[-0.02935723587870598, 0.022572703659534454, -0.010296245105564594, -0.03565344586968422, 0.010964639484882355, 0.014116673730313778, 0.02508564107120037, -0.0064797066152095795, 0.011858065612614155, 0.06767700612545013, 0.05249015986919403, 0.025064688175916672, 0.014964817091822624, 0.04103029519319534, 0.0056918137...
974
974
['Mathias Berglund', 'Tapani Raiko']
1312.6002v3
Contrastive Divergence (CD) and Persistent Contrastive Divergence (PCD) are popular methods for training the weights of Restricted Boltzmann Machines. However, both methods use an approximate method for sampling from the model distribution. As a side effect, these approximations yield significantly different biases and...
Stochastic Gradient Estimate Variance in Contrastive Divergence and Persistent Contrastive Divergence
2,013
http://arxiv.org/pdf/1312.6002v3
Title Stochastic Gradient Estimate Variance Contrastive Divergence Persistent Contrastive Divergence Summary Contrastive Divergence CD Persistent Contrastive Divergence PCD popular method training weight Restricted Boltzmann Machines However method use approximate method sampling model distribution side effect approxim...
[-0.005314758513122797, -0.021308433264493942, -0.029232794418931007, 0.00809037871658802, -0.053116679191589355, -0.007740939036011696, 0.07496026903390884, -0.006683580577373505, -0.059172842651605606, 0.049565389752388, 0.01710551045835018, -0.007842803373932838, 0.052812639623880386, 0.007303531747311354, 0.0267443...
975
975
['Razvan Pascanu', 'Caglar Gulcehre', 'Kyunghyun Cho', 'Yoshua Bengio']
1312.6026v5
In this paper, we explore different ways to extend a recurrent neural network (RNN) to a \textit{deep} RNN. We start by arguing that the concept of depth in an RNN is not as clear as it is in feedforward neural networks. By carefully analyzing and understanding the architecture of an RNN, however, we find three points ...
How to Construct Deep Recurrent Neural Networks
2,013
http://arxiv.org/pdf/1312.6026v5
Title Construct Deep Recurrent Neural Networks Summary paper explore different way extend recurrent neural network RNN textitdeep RNN start arguing concept depth RNN clear feedforward neural network carefully analyzing understanding architecture RNN however find three point RNN may made deeper 1 inputtohidden function ...
[0.012716674245893955, 0.02758781984448433, 0.004509675782173872, 0.07180525362491608, -0.013282503001391888, -0.04989452660083771, 0.03858890011906624, -0.020954087376594543, -0.050997212529182434, 0.02703608199954033, -0.04728904739022255, -0.01823623850941658, 0.02020104043185711, 0.029123356565833092, 0.02196355722...
976
976
['David P. Reichert', 'Thomas Serre']
1312.6115v5
Deep learning has recently led to great successes in tasks such as image recognition (e.g Krizhevsky et al., 2012). However, deep networks are still outmatched by the power and versatility of the brain, perhaps in part due to the richer neuronal computations available to cortical circuits. The challenge is to identify ...
Neuronal Synchrony in Complex-Valued Deep Networks
2,013
http://arxiv.org/pdf/1312.6115v5
Title Neuronal Synchrony ComplexValued Deep Networks Summary Deep learning recently led great success task image recognition eg Krizhevsky et al 2012 However deep network still outmatched power versatility brain perhaps part due richer neuronal computation available cortical circuit challenge identify neuronal mechanis...
[-0.024427782744169235, 0.018502578139305115, -0.01420583389699459, 0.02766118012368679, 0.0035562333650887012, 0.020545832812786102, 0.03273328021168709, -0.011375727131962776, -0.011664996854960918, 0.03170039877295494, -0.03996220603585243, -0.00932775717228651, 0.027115266770124435, 0.05409993976354599, 0.026172310...
977
977
['David Warde-Farley', 'Ian J. Goodfellow', 'Aaron Courville', 'Yoshua Bengio']
1312.6197v2
The recently introduced dropout training criterion for neural networks has been the subject of much attention due to its simplicity and remarkable effectiveness as a regularizer, as well as its interpretation as a training procedure for an exponentially large ensemble of networks that share parameters. In this work we ...
An empirical analysis of dropout in piecewise linear networks
2,013
http://arxiv.org/pdf/1312.6197v2
Title empirical analysis dropout piecewise linear network Summary recently introduced dropout training criterion neural network subject much attention due simplicity remarkable effectiveness regularizer well interpretation training procedure exponentially large ensemble network share parameter work empirically investig...
[-0.030541466549038887, 0.03167891129851341, -0.016241639852523804, 0.0310830045491457, 0.008423184044659138, 0.021112192422151566, 0.08107151091098785, -0.007142371963709593, 0.01009425986558199, 0.003466809866949916, 0.01621749810874462, 0.016501571983098984, 0.002564777387306094, 0.09142162650823593, 0.0281115770339...
978
978
['Ian J. Goodfellow', 'Mehdi Mirza', 'Da Xiao', 'Aaron Courville', 'Yoshua Bengio']
1312.6211v3
Catastrophic forgetting is a problem faced by many machine learning models and algorithms. When trained on one task, then trained on a second task, many machine learning models "forget" how to perform the first task. This is widely believed to be a serious problem for neural networks. Here, we investigate the extent to...
An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks
2,013
http://arxiv.org/pdf/1312.6211v3
Title Empirical Investigation Catastrophic Forgetting GradientBased Neural Networks Summary Catastrophic forgetting problem faced many machine learning model algorithm trained one task trained second task many machine learning model forget perform first task widely believed serious problem neural network investigate ex...
[-0.011497528292238712, 0.013375970534980297, -0.0077837985008955, -0.008122350089251995, 0.013212005607783794, 0.022770186886191368, 0.019457323476672173, -0.012820876203477383, 0.017209367826581, 0.0022365832701325417, -0.0006202684016898274, 0.022808432579040527, -0.017661139369010925, 0.03322286531329155, 0.0187323...
979
979
['Yaroslav Ganin', 'Victor Lempitsky']
1409.7495v2
Top-performing deep architectures are trained on massive amounts of labeled data. In the absence of labeled data for a certain task, domain adaptation often provides an attractive option given that labeled data of similar nature but from a different domain (e.g. synthetic images) are available. Here, we propose a new a...
Unsupervised Domain Adaptation by Backpropagation
2,014
http://arxiv.org/pdf/1409.7495v2
Title Unsupervised Domain Adaptation Backpropagation Summary Topperforming deep architecture trained massive amount labeled data absence labeled data certain task domain adaptation often provides attractive option given labeled data similar nature different domain eg synthetic image available propose new approach domai...
[0.012223862111568451, 0.02730708196759224, -0.029504571110010147, 0.002959083067253232, 0.022374561056494713, 0.013824899680912495, 0.06219928711652756, 0.005808903370052576, -0.025141624733805656, -0.011068616062402725, -0.04348711669445038, 0.010070228949189186, -0.03426011651754379, 0.07055418193340302, -0.00909160...
980
980
['Sherjil Ozair', 'Yoshua Bengio']
1410.0630v1
For discrete data, the likelihood $P(x)$ can be rewritten exactly and parametrized into $P(X = x) = P(X = x | H = f(x)) P(H = f(x))$ if $P(X | H)$ has enough capacity to put no probability mass on any $x'$ for which $f(x')\neq f(x)$, where $f(\cdot)$ is a deterministic discrete function. The log of the first factor giv...
Deep Directed Generative Autoencoders
2,014
http://arxiv.org/pdf/1410.0630v1
Title Deep Directed Generative Autoencoders Summary discrete data likelihood Px rewritten exactly parametrized PX x PX x H fx PH fx PX H enough capacity put probability mass x fxneq fx fcdot deterministic discrete function log first factor give rise loglikelihood reconstruction error autoencoder fcdot encoder PXH proba...
[-0.04753543436527252, 0.08173410594463348, -0.012823665514588356, 0.012111885473132133, 0.026507612317800522, -0.01866910606622696, 0.040435802191495895, -0.017728904262185097, -0.07300161570310593, 0.029092561453580856, 0.011009586974978447, 0.005185193847864866, 0.002752364845946431, 0.08532621711492538, 0.061463844...
981
981
['Pankaj Mehta', 'David J. Schwab']
1410.3831v1
Deep learning is a broad set of techniques that uses multiple layers of representation to automatically learn relevant features directly from structured data. Recently, such techniques have yielded record-breaking results on a diverse set of difficult machine learning tasks in computer vision, speech recognition, and n...
An exact mapping between the Variational Renormalization Group and Deep Learning
2,014
http://arxiv.org/pdf/1410.3831v1
Title exact mapping Variational Renormalization Group Deep Learning Summary Deep learning broad set technique us multiple layer representation automatically learn relevant feature directly structured data Recently technique yielded recordbreaking result diverse set difficult machine learning task computer vision speech...
[-0.0023762525524944067, 0.020458418875932693, -0.03103753924369812, 0.032984085381031036, -0.006900325883179903, 0.008116996847093105, 0.007856132462620735, -0.0002189859951613471, -0.06804229319095612, 0.02625308558344841, -0.01627558283507824, -0.03649413585662842, -0.012955756857991219, 0.0626741424202919, 0.011091...
982
982
['Erte Pan', 'Zhu Han']
1410.4599v2
In this paper, we present an infinite hierarchical non-parametric Bayesian model to extract the hidden factors over observed data, where the number of hidden factors for each layer is unknown and can be potentially infinite. Moreover, the number of layers can also be infinite. We construct the model structure that allo...
Non-parametric Bayesian Learning with Deep Learning Structure and Its Applications in Wireless Networks
2,014
http://arxiv.org/pdf/1410.4599v2
Title Nonparametric Bayesian Learning Deep Learning Structure Applications Wireless Networks Summary paper present infinite hierarchical nonparametric Bayesian model extract hidden factor observed data number hidden factor layer unknown potentially infinite Moreover number layer also infinite construct model structure ...
[-0.06093990057706833, 0.016605103388428688, -0.0024285053368657827, -0.014800445176661015, -0.010646031238138676, -0.004425301216542721, 0.03849741071462631, -0.03221261501312256, -0.05683831870555878, 0.002845281967893243, -0.011117312125861645, 0.017581436783075333, 0.023697085678577423, 0.03657323867082596, 0.00429...
983
983
['Daniel Povey', 'Xiaohui Zhang', 'Sanjeev Khudanpur']
1410.7455v8
We describe the neural-network training framework used in the Kaldi speech recognition toolkit, which is geared towards training DNNs with large amounts of training data using multiple GPU-equipped or multi-core machines. In order to be as hardware-agnostic as possible, we needed a way to use multiple machines without ...
Parallel training of DNNs with Natural Gradient and Parameter Averaging
2,014
http://arxiv.org/pdf/1410.7455v8
Title Parallel training DNNs Natural Gradient Parameter Averaging Summary describe neuralnetwork training framework used Kaldi speech recognition toolkit geared towards training DNNs large amount training data using multiple GPUequipped multicore machine order hardwareagnostic possible needed way use multiple machine w...
[-0.02147987298667431, 0.03240162879228592, 0.011054708622395992, 0.020620791241526604, -0.007865452207624912, 0.013133641332387924, 0.05932673066854477, -0.039054032415151596, -0.04459226876497269, 0.02164766564965248, -0.03262799605727196, -0.00869940035045147, 0.04856467247009277, -0.004756217356771231, 0.0011192006...
984
984
['Jan Chorowski', 'Dzmitry Bahdanau', 'Kyunghyun Cho', 'Yoshua Bengio']
1412.1602v1
We replace the Hidden Markov Model (HMM) which is traditionally used in in continuous speech recognition with a bi-directional recurrent neural network encoder coupled to a recurrent neural network decoder that directly emits a stream of phonemes. The alignment between the input and output sequences is established usin...
End-to-end Continuous Speech Recognition using Attention-based Recurrent NN: First Results
2,014
http://arxiv.org/pdf/1412.1602v1
Title Endtoend Continuous Speech Recognition using Attentionbased Recurrent NN First Results Summary replace Hidden Markov Model HMM traditionally used continuous speech recognition bidirectional recurrent neural network encoder coupled recurrent neural network decoder directly emits stream phoneme alignment input outp...
[-0.03826582059264183, 0.04862799495458603, 0.02661972865462303, 0.037740353494882584, 0.005929560400545597, -0.0165259912610054, 0.01713675633072853, -0.03426314890384674, -0.035356104373931885, -0.00047799330786801875, -0.037393536418676376, -0.008781025186181068, 0.06878233700990677, 0.05213642120361328, 0.019133005...
985
985
['Hanie Sedghi', 'Anima Anandkumar']
1412.2693v4
We provide novel guaranteed approaches for training feedforward neural networks with sparse connectivity. We leverage on the techniques developed previously for learning linear networks and show that they can also be effectively adopted to learn non-linear networks. We operate on the moments involving label and the sco...
Provable Methods for Training Neural Networks with Sparse Connectivity
2,014
http://arxiv.org/pdf/1412.2693v4
Title Provable Methods Training Neural Networks Sparse Connectivity Summary provide novel guaranteed approach training feedforward neural network sparse connectivity leverage technique developed previously learning linear network show also effectively adopted learn nonlinear network operate moment involving label score...
[-0.02425442263484001, 0.05809585750102997, 0.002752972301095724, 0.037369973957538605, 0.04178553819656372, -0.06789786368608475, 0.004382973071187735, -0.0024964474141597748, -0.007964118383824825, 0.026224713772535324, 0.0039517381228506565, -0.0015477631241083145, 0.028708918020129204, 0.06455957889556885, 0.028444...
986
986
['Hana Ajakan', 'Pascal Germain', 'Hugo Larochelle', 'François Laviolette', 'Mario Marchand']
1412.4446v2
We introduce a new representation learning algorithm suited to the context of domain adaptation, in which data at training and test time come from similar but different distributions. Our algorithm is directly inspired by theory on domain adaptation suggesting that, for effective domain transfer to be achieved, predict...
Domain-Adversarial Neural Networks
2,014
http://arxiv.org/pdf/1412.4446v2
Title DomainAdversarial Neural Networks Summary introduce new representation learning algorithm suited context domain adaptation data training test time come similar different distribution algorithm directly inspired theory domain adaptation suggesting effective domain transfer achieved prediction must made based data ...
[0.006073937751352787, 0.024887021631002426, -0.03739439323544502, 0.041518259793519974, 0.010741428472101688, -0.015602017752826214, 0.0417422391474247, -0.008342403918504715, -0.01444221194833517, -0.047414060682058334, -0.05214863643050194, -0.007189407013356686, -0.0120255621150136, 0.06032171845436096, 0.004629259...
987
987
['Philip Bachman', 'Ouais Alsharif', 'Doina Precup']
1412.4864v1
We formalize the notion of a pseudo-ensemble, a (possibly infinite) collection of child models spawned from a parent model by perturbing it according to some noise process. E.g., dropout (Hinton et. al, 2012) in a deep neural network trains a pseudo-ensemble of child subnetworks generated by randomly masking nodes in t...
Learning with Pseudo-Ensembles
2,014
http://arxiv.org/pdf/1412.4864v1
Title Learning PseudoEnsembles Summary formalize notion pseudoensemble possibly infinite collection child model spawned parent model perturbing according noise process Eg dropout Hinton et al 2012 deep neural network train pseudoensemble child subnetworks generated randomly masking node parent network present novel reg...
[-0.009160557761788368, 0.05967608839273453, -0.028614666312932968, 0.03325432538986206, -0.011192547157406807, -0.007061092182993889, 0.003979053348302841, 0.022278467193245888, -0.007695604115724564, 0.001157507300376892, 0.0103920828551054, 0.0007392219849862158, -0.046328116208314896, 0.047495149075984955, -0.00245...
988
988
['David Sussillo', 'L. F. Abbott']
1412.6558v3
Training very deep networks is an important open problem in machine learning. One of many difficulties is that the norm of the back-propagated error gradient can grow or decay exponentially. Here we show that training very deep feed-forward networks (FFNs) is not as difficult as previously thought. Unlike when back-pro...
Random Walk Initialization for Training Very Deep Feedforward Networks
2,014
http://arxiv.org/pdf/1412.6558v3
Title Random Walk Initialization Training Deep Feedforward Networks Summary Training deep network important open problem machine learning One many difficulty norm backpropagated error gradient grow decay exponentially show training deep feedforward network FFNs difficult previously thought Unlike backpropagation applie...
[-0.0044240802526474, 0.0069373114965856075, -0.013173596933484077, -0.0073425062000751495, 0.009337766095995903, -0.00873428676277399, 0.028067974373698235, -0.035599641501903534, 0.0035190260969102383, 0.02239438146352768, 0.04211749881505966, -0.011814677156507969, 0.012025756761431694, 0.07086333632469177, 0.023403...
989
989
['Otto Fabius', 'Joost R. van Amersfoort']
1412.6581v6
In this paper we propose a model that combines the strengths of RNNs and SGVB: the Variational Recurrent Auto-Encoder (VRAE). Such a model can be used for efficient, large scale unsupervised learning on time series data, mapping the time series data to a latent vector representation. The model is generative, such that ...
Variational Recurrent Auto-Encoders
2,014
http://arxiv.org/pdf/1412.6581v6
Title Variational Recurrent AutoEncoders Summary paper propose model combine strength RNNs SGVB Variational Recurrent AutoEncoder VRAE model used efficient large scale unsupervised learning time series data mapping time series data latent vector representation model generative data generated sample latent space importa...
[-0.018449978902935982, 0.06967470049858093, -0.02229575254023075, 0.00022500869818031788, -0.01353391632437706, 0.015410826541483402, 0.01675359718501568, -0.018347375094890594, -0.03326942399144173, -0.0021442759316414595, 0.020723307505249977, -0.017815684899687767, 0.03199680149555206, 0.0761215090751648, 0.0360033...
990
990
['Jan Rudy', 'Weiguang Ding', 'Daniel Jiwoong Im', 'Graham W. Taylor']
1412.6630v2
Regularization is essential when training large neural networks. As deep neural networks can be mathematically interpreted as universal function approximators, they are effective at memorizing sampling noise in the training data. This results in poor generalization to unseen data. Therefore, it is no surprise that a ne...
Neural Network Regularization via Robust Weight Factorization
2,014
http://arxiv.org/pdf/1412.6630v2
Title Neural Network Regularization via Robust Weight Factorization Summary Regularization essential training large neural network deep neural network mathematically interpreted universal function approximators effective memorizing sampling noise training data result poor generalization unseen data Therefore surprise n...
[-0.019928568974137306, 0.06103457137942314, -0.019916892051696777, 0.03545447811484337, 0.054630864411592484, -0.026063794270157814, 0.03763958811759949, -0.020228462293744087, -0.031058156862854958, 0.05151055380702019, 0.014162488281726837, -0.0016411741962656379, 0.004072746727615595, 0.05240050330758095, 0.0318381...
991
991
['Shin-ichi Maeda']
1412.7003v3
Dropout is one of the key techniques to prevent the learning from overfitting. It is explained that dropout works as a kind of modified L2 regularization. Here, we shed light on the dropout from Bayesian standpoint. Bayesian interpretation enables us to optimize the dropout rate, which is beneficial for learning of wei...
A Bayesian encourages dropout
2,014
http://arxiv.org/pdf/1412.7003v3
Title Bayesian encourages dropout Summary Dropout one key technique prevent learning overfitting explained dropout work kind modified L2 regularization shed light dropout Bayesian standpoint Bayesian interpretation enables u optimize dropout rate beneficial learning weight parameter prediction learning experiment resul...
[-0.011000306345522404, 0.04186517000198364, -0.02787579782307148, -0.0035959735978394747, -0.01786748692393303, 0.014211051166057587, 0.04555059224367142, 0.02868657559156418, -0.015029782429337502, -0.0038209096528589725, 0.04330866038799286, 0.05958765745162964, 0.0038028869312256575, 0.05666129291057587, -0.0077756...
992
992
['Zichao Yang', 'Marcin Moczulski', 'Misha Denil', 'Nando de Freitas', 'Alex Smola', 'Le Song', 'Ziyu Wang']
1412.7149v4
The fully connected layers of a deep convolutional neural network typically contain over 90% of the network parameters, and consume the majority of the memory required to store the network parameters. Reducing the number of parameters while preserving essentially the same predictive performance is critically important ...
Deep Fried Convnets
2,014
http://arxiv.org/pdf/1412.7149v4
Title Deep Fried Convnets Summary fully connected layer deep convolutional neural network typically contain 90 network parameter consume majority memory required store network parameter Reducing number parameter preserving essentially predictive performance critically important operating deep neural network memory cons...
[0.003953681327402592, 0.04814979061484337, -0.007932237349450588, 0.0726790726184845, 0.030183227732777596, -0.013543183915317059, 0.023007377982139587, 0.00977381318807602, -0.007258429657667875, -0.01303425244987011, -0.04898957535624504, 0.016132162883877754, 0.009048628620803356, 0.06714611500501633, 0.02202516980...
993
993
['Antti Rasmus', 'Harri Valpola', 'Tapani Raiko']
1504.08215v1
We show how a deep denoising autoencoder with lateral connections can be used as an auxiliary unsupervised learning task to support supervised learning. The proposed model is trained to minimize simultaneously the sum of supervised and unsupervised cost functions by back-propagation, avoiding the need for layer-wise pr...
Lateral Connections in Denoising Autoencoders Support Supervised Learning
2,015
http://arxiv.org/pdf/1504.08215v1
Title Lateral Connections Denoising Autoencoders Support Supervised Learning Summary show deep denoising autoencoder lateral connection used auxiliary unsupervised learning task support supervised learning proposed model trained minimize simultaneously sum supervised unsupervised cost function backpropagation avoiding ...
[-0.04022599756717682, 0.03306163102388382, -0.04414362832903862, 0.050157252699136734, 0.027704793959856033, 0.00064628547988832, 0.022003471851348877, -0.0456169918179512, -0.013479175046086311, 0.006852654740214348, -0.051992353051900864, 0.017777154222130775, 0.037417296320199966, 0.059839680790901184, 0.0396613441...
994
994
['Raja Giryes', 'Guillermo Sapiro', 'Alex M. Bronstein']
1504.08291v5
Three important properties of a classification machinery are: (i) the system preserves the core information of the input data; (ii) the training examples convey information about unseen data; and (iii) the system is able to treat differently points from different classes. In this work we show that these fundamental pro...
Deep Neural Networks with Random Gaussian Weights: A Universal Classification Strategy?
2,015
http://arxiv.org/pdf/1504.08291v5
Title Deep Neural Networks Random Gaussian Weights Universal Classification Strategy Summary Three important property classification machinery system preserve core information input data ii training example convey information unseen data iii system able treat differently point different class work show fundamental prop...
[-0.01569700427353382, 0.03660830110311508, -0.02110263891518116, 0.015260525979101658, -0.02043360099196434, 0.0010939789935946465, 0.0314251184463501, 0.0053098187781870365, -0.01996506005525589, 0.006346167530864477, 0.031633105129003525, -0.007351071573793888, 0.022093666717410088, 0.03891381621360779, 0.0132181607...
995
995
['Zhiguang Wang', 'Tim Oates']
1506.00327v1
Inspired by recent successes of deep learning in computer vision, we propose a novel framework for encoding time series as different types of images, namely, Gramian Angular Summation/Difference Fields (GASF/GADF) and Markov Transition Fields (MTF). This enables the use of techniques from computer vision for time serie...
Imaging Time-Series to Improve Classification and Imputation
2,015
http://arxiv.org/pdf/1506.00327v1
Title Imaging TimeSeries Improve Classification Imputation Summary Inspired recent success deep learning computer vision propose novel framework encoding time series different type image namely Gramian Angular SummationDifference Fields GASFGADF Markov Transition Fields MTF enables use technique computer vision time se...
[-0.03319625183939934, 0.09899730235338211, -0.005162252578884363, 0.042760130017995834, 0.0001874628651421517, 0.036478545516729355, 0.049345582723617554, 0.04922117292881012, 0.0013519965577870607, 0.03677152097225189, 0.034278277307748795, -0.02306956611573696, 0.03465266898274422, 0.049860235303640366, 0.0075896903...
996
996
['Bart van Merriënboer', 'Dzmitry Bahdanau', 'Vincent Dumoulin', 'Dmitriy Serdyuk', 'David Warde-Farley', 'Jan Chorowski', 'Yoshua Bengio']
1506.00619v1
We introduce two Python frameworks to train neural networks on large datasets: Blocks and Fuel. Blocks is based on Theano, a linear algebra compiler with CUDA-support. It facilitates the training of complex neural network models by providing parametrized Theano operations, attaching metadata to Theano's symbolic comput...
Blocks and Fuel: Frameworks for deep learning
2,015
http://arxiv.org/pdf/1506.00619v1
Title Blocks Fuel Frameworks deep learning Summary introduce two Python framework train neural network large datasets Blocks Fuel Blocks based Theano linear algebra compiler CUDAsupport facilitates training complex neural network model providing parametrized Theano operation attaching metadata Theanos symbolic computat...
[-0.023373784497380257, 0.03020104207098484, -0.039356790482997894, 0.010949266143143177, 0.0060263327322900295, 0.03363582491874695, 0.07424229383468628, 0.018225261941552162, -0.019602276384830475, -0.015069367364048958, 0.038894981145858765, 0.006550794467329979, -0.002780164824798703, 0.09700920432806015, 0.0244682...
997
997
['Zhiguang Wang', 'Tim Oates', 'James Lo']
1506.02690v3
This paper proposes a set of new error criteria and learning approaches, Adaptive Normalized Risk-Averting Training (ANRAT), to attack the non-convex optimization problem in training deep neural networks (DNNs). Theoretically, we demonstrate its effectiveness on global and local convexity lower-bounded by the standard ...
Adaptive Normalized Risk-Averting Training For Deep Neural Networks
2,015
http://arxiv.org/pdf/1506.02690v3
Title Adaptive Normalized RiskAverting Training Deep Neural Networks Summary paper proposes set new error criterion learning approach Adaptive Normalized RiskAverting Training ANRAT attack nonconvex optimization problem training deep neural network DNNs Theoretically demonstrate effectiveness global local convexity low...
[-0.03892209753394127, 0.02396414242684841, -0.007979538291692734, 0.04277722164988518, 0.02514675445854664, 0.005754395853728056, 0.03764493763446808, -0.014654085971415043, -0.04229920357465744, 0.021381240338087082, -0.03835045546293259, 0.01774756796658039, -0.02433912083506584, 0.010545829311013222, 0.037587128579...
998
998
['Marylou Gabrié', 'Eric W. Tramel', 'Florent Krzakala']
1506.02914v2
Restricted Boltzmann machines are undirected neural networks which have been shown to be effective in many applications, including serving as initializations for training deep multi-layer neural networks. One of the main reasons for their success is the existence of efficient and practical stochastic algorithms, such a...
Training Restricted Boltzmann Machines via the Thouless-Anderson-Palmer Free Energy
2,015
http://arxiv.org/pdf/1506.02914v2
Title Training Restricted Boltzmann Machines via ThoulessAndersonPalmer Free Energy Summary Restricted Boltzmann machine undirected neural network shown effective many application including serving initialization training deep multilayer neural network One main reason success existence efficient practical stochastic al...
[-0.017963889986276627, 0.01140306144952774, -0.014971827156841755, 0.04444911330938339, -0.01435950305312872, 0.0023342478089034557, 0.062007542699575424, -0.005196407437324524, -0.01400814950466156, -0.0021559600718319416, -0.030290434136986732, -0.00036210299003869295, 0.024807089939713478, 0.025061191990971565, 0.0...
999
999
['Oriol Vinyals', 'Meire Fortunato', 'Navdeep Jaitly']
1506.03134v2
We introduce a new neural architecture to learn the conditional probability of an output sequence with elements that are discrete tokens corresponding to positions in an input sequence. Such problems cannot be trivially addressed by existent approaches such as sequence-to-sequence and Neural Turing Machines, because th...
Pointer Networks
2,015
http://arxiv.org/pdf/1506.03134v2
Title Pointer Networks Summary introduce new neural architecture learn conditional probability output sequence element discrete token corresponding position input sequence problem cannot trivially addressed existent approach sequencetosequence Neural Turing Machines number target class step output depends length input ...
[0.003444165922701359, 0.021055107936263084, -0.022053953260183334, 0.05184723064303398, -0.019682303071022034, -0.03767344728112221, -0.007297187112271786, -0.029132695868611336, -0.04402274638414383, 0.011482217349112034, 0.03488033637404442, 0.014378183521330357, 0.003138013882562518, 0.055737096816301346, 0.0540490...