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
800
800
['Kamil Rocki']
1603.08262v1
There exists a theory of a single general-purpose learning algorithm which could explain the principles of its operation. This theory assumes that the brain has some initial rough architecture, a small library of simple innate circuits which are prewired at birth and proposes that all significant mental algorithms can ...
Towards Machine Intelligence
2,016
http://arxiv.org/pdf/1603.08262v1
Title Towards Machine Intelligence Summary exists theory single generalpurpose learning algorithm could explain principle operation theory assumes brain initial rough architecture small library simple innate circuit prewired birth proposes significant mental algorithm learned Given current understanding observation pap...
[-0.017351990565657616, 0.024881742894649506, -0.04097719117999077, -0.012972159311175346, -0.008411930873990059, 0.01303438562899828, 0.0025483272038400173, -0.003180443774908781, 0.003136167535558343, 0.001992369769141078, 0.021276718005537987, 0.034689806401729584, 0.0003844456223305315, 0.04120863974094391, 0.06122...
801
801
['Aravind S. Lakshminarayanan', 'Sahil Sharma', 'Balaraman Ravindran']
1605.05365v2
Deep Reinforcement Learning methods have achieved state of the art performance in learning control policies for the games in the Atari 2600 domain. One of the important parameters in the Arcade Learning Environment (ALE) is the frame skip rate. It decides the granularity at which agents can control game play. A frame s...
Dynamic Frame skip Deep Q Network
2,016
http://arxiv.org/pdf/1605.05365v2
Title Dynamic Frame skip Deep Q Network Summary Deep Reinforcement Learning method achieved state art performance learning control policy game Atari 2600 domain One important parameter Arcade Learning Environment ALE frame skip rate decides granularity agent control game play frame skip value k allows agent repeat sele...
[-0.020967982709407806, 0.025493472814559937, -0.03489236533641815, -0.02994390018284321, 0.0019058444304391742, 0.0023948538582772017, 0.02895328775048256, -0.0014554777881130576, -0.05222182348370552, 0.022573666647076607, -0.02657044120132923, 0.008481152355670929, -0.027061982080340385, 0.10703840106725693, 0.01127...
802
802
['Matko Bošnjak', 'Tim Rocktäschel', 'Jason Naradowsky', 'Sebastian Riedel']
1605.06640v3
Given that in practice training data is scarce for all but a small set of problems, a core question is how to incorporate prior knowledge into a model. In this paper, we consider the case of prior procedural knowledge for neural networks, such as knowing how a program should traverse a sequence, but not what local acti...
Programming with a Differentiable Forth Interpreter
2,016
http://arxiv.org/pdf/1605.06640v3
Title Programming Differentiable Forth Interpreter Summary Given practice training data scarce small set problem core question incorporate prior knowledge model paper consider case prior procedural knowledge neural network knowing program traverse sequence local action performed step end present endtoend differentiable...
[0.015784751623868942, 0.02968372032046318, -0.02548852376639843, 0.03009185940027237, -0.0074228448793292046, 0.014865677803754807, -0.004632242955267429, -0.026738736778497696, -0.010036690160632133, -0.0015159444883465767, -0.00042609203956089914, 0.030986011028289795, 0.02082044444978237, 0.09317515045404434, 0.002...
803
803
['Luka Crnkovic-Friis', 'Louise Crnkovic-Friis']
1605.06921v1
Recent advances in deep learning have enabled the extraction of high-level features from raw sensor data which has opened up new possibilities in many different fields, including computer generated choreography. In this paper we present a system chor-rnn for generating novel choreographic material in the nuanced choreo...
Generative Choreography using Deep Learning
2,016
http://arxiv.org/pdf/1605.06921v1
Title Generative Choreography using Deep Learning Summary Recent advance deep learning enabled extraction highlevel feature raw sensor data opened new possibility many different field including computer generated choreography paper present system chorrnn generating novel choreographic material nuanced choreographic lan...
[-0.014845410361886024, 0.01873498037457466, -0.0018858957337215543, 0.01969604566693306, -0.0060719698667526245, -0.0038841378409415483, 0.010183590464293957, 0.007069035898894072, -0.02876858413219452, 0.01212883461266756, -0.04341796785593033, -0.028224367648363113, 0.0031514095608145, 0.08033016324043274, 0.0364874...
804
804
['Luciano Serafini', "Artur d'Avila Garcez"]
1606.04422v2
We propose Logic Tensor Networks: a uniform framework for integrating automatic learning and reasoning. A logic formalism called Real Logic is defined on a first-order language whereby formulas have truth-value in the interval [0,1] and semantics defined concretely on the domain of real numbers. Logical constants are i...
Logic Tensor Networks: Deep Learning and Logical Reasoning from Data and Knowledge
2,016
http://arxiv.org/pdf/1606.04422v2
Title Logic Tensor Networks Deep Learning Logical Reasoning Data Knowledge Summary propose Logic Tensor Networks uniform framework integrating automatic learning reasoning logic formalism called Real Logic defined firstorder language whereby formula truthvalue interval 01 semantics defined concretely domain real number...
[-0.005493875592947006, 0.061042167246341705, -0.020459052175283432, 0.0005539292469620705, -0.03672215715050697, 0.006175587419420481, 0.04792485013604164, 0.003593750298023224, 0.014896784909069538, -0.02431536838412285, 0.050967440009117126, 0.044971566647291183, -0.0448899008333683, 0.0772092193365097, -0.027581647...
805
805
['Randal S. Olson', 'Jason H. Moore']
1607.08878v1
As data science continues to grow in popularity, there will be an increasing need to make data science tools more scalable, flexible, and accessible. In particular, automated machine learning (AutoML) systems seek to automate the process of designing and optimizing machine learning pipelines. In this chapter, we presen...
Identifying and Harnessing the Building Blocks of Machine Learning Pipelines for Sensible Initialization of a Data Science Automation Tool
2,016
http://arxiv.org/pdf/1607.08878v1
Title Identifying Harnessing Building Blocks Machine Learning Pipelines Sensible Initialization Data Science Automation Tool Summary data science continues grow popularity increasing need make data science tool scalable flexible accessible particular automated machine learning AutoML system seek automate process design...
[-0.006545540411025286, 0.025100000202655792, -0.06344747543334961, -0.031780652701854706, -0.002485371194779873, 0.033455368131399155, 0.04328327625989914, -0.003430407028645277, 0.03076452575623989, -0.03438777104020119, 0.051863860338926315, 0.027494585141539574, -0.0013480602065101266, 0.124763622879982, 0.01202288...
806
806
['Karan K. Budhraja', 'Tim Oates']
1608.02971v1
The problem of Learning from Demonstration is targeted at learning to perform tasks based on observed examples. One approach to Learning from Demonstration is Inverse Reinforcement Learning, in which actions are observed to infer rewards. This work combines a feature based state evaluation approach to Inverse Reinforce...
Neuroevolution-Based Inverse Reinforcement Learning
2,016
http://arxiv.org/pdf/1608.02971v1
Title NeuroevolutionBased Inverse Reinforcement Learning Summary problem Learning Demonstration targeted learning perform task based observed example One approach Learning Demonstration Inverse Reinforcement Learning action observed infer reward work combine feature based state evaluation approach Inverse Reinforcement...
[0.015042547136545181, 0.004591308068484068, -0.005603528581559658, -0.045446816831827164, -0.016664888709783554, -0.04285820201039314, -0.03574316203594208, 0.02535259909927845, -0.057598140090703964, 0.013222440145909786, -0.04284298047423363, 0.04760914668440819, -0.013777272775769234, 0.0709560215473175, 0.06078116...
807
807
['Alexander L. Gaunt', 'Marc Brockschmidt', 'Rishabh Singh', 'Nate Kushman', 'Pushmeet Kohli', 'Jonathan Taylor', 'Daniel Tarlow']
1608.04428v1
We study machine learning formulations of inductive program synthesis; given input-output examples, we try to synthesize source code that maps inputs to corresponding outputs. Our aims are to develop new machine learning approaches based on neural networks and graphical models, and to understand the capabilities of mac...
TerpreT: A Probabilistic Programming Language for Program Induction
2,016
http://arxiv.org/pdf/1608.04428v1
Title TerpreT Probabilistic Programming Language Program Induction Summary study machine learning formulation inductive program synthesis given inputoutput example try synthesize source code map input corresponding output aim develop new machine learning approach based neural network graphical model understand capabili...
[-0.01722017489373684, 0.017943255603313446, -0.027007954195141792, -0.010089694522321224, -0.024836236611008644, -0.023966439068317413, -0.04823090881109238, -0.02402658388018608, 0.0008894599159248173, -0.03042583353817463, 0.08167152106761932, 0.08090608566999435, 0.025328921154141426, 0.10649725049734116, 0.0158349...
808
808
['Rajasekar Venkatesan', 'Meng Joo Er']
1608.08435v1
In this paper, an Extreme Learning Machine (ELM) based technique for Multi-label classification problems is proposed and discussed. In multi-label classification, each of the input data samples belongs to one or more than one class labels. The traditional binary and multi-class classification problems are the subset of...
Multi-Label Classification Method Based on Extreme Learning Machines
2,016
http://arxiv.org/pdf/1608.08435v1
Title MultiLabel Classification Method Based Extreme Learning Machines Summary paper Extreme Learning Machine ELM based technique Multilabel classification problem proposed discussed multilabel classification input data sample belongs one one class label traditional binary multiclass classification problem subset multi...
[0.03651493787765503, -0.03894156590104103, -0.02149415761232376, -0.016799520701169968, 0.014798006042838097, 0.04487765580415726, 0.0054421694949269295, 0.023157594725489616, 0.021656595170497894, -0.04483086243271828, -0.013578305952250957, 0.056093357503414154, -0.013931783847510815, 0.01588790863752365, 0.01021040...
809
809
['Rajasekar Venkatesan', 'Meng Joo Er', 'Shiqian Wu', 'Mahardhika Pratama']
1608.08905v1
In this paper, a novel extreme learning machine based online multi-label classifier for real-time data streams is proposed. Multi-label classification is one of the actively researched machine learning paradigm that has gained much attention in the recent years due to its rapidly increasing real world applications. In ...
A Novel Online Real-time Classifier for Multi-label Data Streams
2,016
http://arxiv.org/pdf/1608.08905v1
Title Novel Online Realtime Classifier Multilabel Data Streams Summary paper novel extreme learning machine based online multilabel classifier realtime data stream proposed Multilabel classification one actively researched machine learning paradigm gained much attention recent year due rapidly increasing real world app...
[-0.006298888940364122, 0.0011950135231018066, -0.015283727087080479, -0.010427319444715977, 0.01822822168469429, 0.033135946840047836, 0.03005293756723404, -0.00841550249606371, 0.024135859683156013, -0.03136824443936348, 0.005537543911486864, 0.0399785116314888, -0.029923059046268463, 0.07971092313528061, 0.021839916...
810
810
['Rajasekar Venkatesan', 'Meng Joo Er']
1609.00085v2
In this paper, a progressive learning technique for multi-class classification is proposed. This newly developed learning technique is independent of the number of class constraints and it can learn new classes while still retaining the knowledge of previous classes. Whenever a new class (non-native to the knowledge le...
A Novel Progressive Learning Technique for Multi-class Classification
2,016
http://arxiv.org/pdf/1609.00085v2
Title Novel Progressive Learning Technique Multiclass Classification Summary paper progressive learning technique multiclass classification proposed newly developed learning technique independent number class constraint learn new class still retaining knowledge previous class Whenever new class nonnative knowledge lear...
[-0.009050623513758183, 0.07899542897939682, -0.013789976015686989, -0.0008588870405219495, -0.010595673695206642, -0.0043508633971214294, 0.03322456777095795, -0.007411185186356306, -0.022002218291163445, -0.026269644498825073, 0.01577657274901867, 0.0244437325745821, 0.007352571468800306, -0.005320791155099869, 0.016...
811
811
['Rajasekar Venkatesan', 'Meng Joo Er', 'Mihika Dave', 'Mahardhika Pratama', 'Shiqian Wu']
1609.00086v1
In this paper, a high-speed online neural network classifier based on extreme learning machines for multi-label classification is proposed. In multi-label classification, each of the input data sample belongs to one or more than one of the target labels. The traditional binary and multi-class classification where each ...
A novel online multi-label classifier for high-speed streaming data applications
2,016
http://arxiv.org/pdf/1609.00086v1
Title novel online multilabel classifier highspeed streaming data application Summary paper highspeed online neural network classifier based extreme learning machine multilabel classification proposed multilabel classification input data sample belongs one one target label traditional binary multiclass classification s...
[-0.0017909012967720628, 0.011116793379187584, -0.012937987223267555, -0.017560698091983795, 0.01685900054872036, 0.02884375862777233, 0.020700978115200996, 0.016809318214654922, 0.024953285232186317, -0.033660922199487686, -0.008609489537775517, 0.03520015627145767, -0.02639882266521454, 0.08068560063838959, 0.0044535...
812
812
['Hande Alemdar', 'Vincent Leroy', 'Adrien Prost-Boucle', 'Frédéric Pétrot']
1609.00222v2
The computation and storage requirements for Deep Neural Networks (DNNs) are usually high. This issue limits their deployability on ubiquitous computing devices such as smart phones, wearables and autonomous drones. In this paper, we propose ternary neural networks (TNNs) in order to make deep learning more resource-ef...
Ternary Neural Networks for Resource-Efficient AI Applications
2,016
http://arxiv.org/pdf/1609.00222v2
Title Ternary Neural Networks ResourceEfficient AI Applications Summary computation storage requirement Deep Neural Networks DNNs usually high issue limit deployability ubiquitous computing device smart phone wearable autonomous drone paper propose ternary neural network TNNs order make deep learning resourceefficient ...
[0.007484246511012316, 0.03200938552618027, -0.02053963765501976, 0.029234547168016434, -0.0013657851377502084, -0.05449562519788742, 0.06825247406959534, -0.017961448058485985, -0.06530588120222092, -0.04260271415114403, 0.01697932556271553, -0.004746080841869116, -0.02053097076714039, 0.04266133904457092, 0.057883877...
813
813
['Navid Kardan', 'Kenneth O. Stanley']
1609.02226v3
Though deep learning has pushed the boundaries of classification forward, in recent years hints of the limits of standard classification have begun to emerge. Problems such as fooling, adding new classes over time, and the need to retrain learning models only for small changes to the original problem all point to a pot...
Fitted Learning: Models with Awareness of their Limits
2,016
http://arxiv.org/pdf/1609.02226v3
Title Fitted Learning Models Awareness Limits Summary Though deep learning pushed boundary classification forward recent year hint limit standard classification begun emerge Problems fooling adding new class time need retrain learning model small change original problem point potential shortcoming classic classificatio...
[0.004553238861262798, 0.023862645030021667, -0.02515711449086666, 0.015982070937752724, 0.004415976349264383, 3.2951364119071513e-05, 0.02864004857838154, 0.004825477488338947, -0.037041936069726944, -0.014320683665573597, 0.0034049013629555702, 0.026574747636914253, -0.010187797248363495, 0.08596868813037872, -0.0012...
814
814
['Thomas Miconi']
1609.02228v2
Hebbian plasticity is a powerful principle that allows biological brains to learn from their lifetime experience. By contrast, artificial neural networks trained with backpropagation generally have fixed connection weights that do not change once training is complete. While recent methods can endow neural networks with...
Learning to learn with backpropagation of Hebbian plasticity
2,016
http://arxiv.org/pdf/1609.02228v2
Title Learning learn backpropagation Hebbian plasticity Summary Hebbian plasticity powerful principle allows biological brain learn lifetime experience contrast artificial neural network trained backpropagation generally fixed connection weight change training complete recent method endow neural network longterm memory...
[-0.005512428469955921, 0.006936597637832165, -0.031196948140859604, -0.028671780601143837, 0.01135297678411007, -0.021965527907013893, 0.037528522312641144, 0.0007433979772031307, -0.019423458725214005, 0.037506431341171265, -0.016754431650042534, 0.021670395508408546, 0.027609504759311676, 0.048013459891080856, 0.020...
815
815
['Lana Sinapayen', 'Atsushi Masumori', 'Takashi Ikegami']
1609.07706v1
Learning based on networks of real neurons, and by extension biologically inspired models of neural networks, has yet to find general learning rules leading to widespread applications. In this paper, we argue for the existence of a principle allowing to steer the dynamics of a biologically inspired neural network. Usin...
Learning by Stimulation Avoidance: A Principle to Control Spiking Neural Networks Dynamics
2,016
http://arxiv.org/pdf/1609.07706v1
Title Learning Stimulation Avoidance Principle Control Spiking Neural Networks Dynamics Summary Learning based network real neuron extension biologically inspired model neural network yet find general learning rule leading widespread application paper argue existence principle allowing steer dynamic biologically inspir...
[-0.018463438376784325, -0.03528013452887535, -0.028859732672572136, -0.003135600360110402, 0.010782533325254917, -0.007587278727442026, -0.0055375006049871445, -0.003981065936386585, 0.01698615401983261, 0.027135254815220833, -0.036468978971242905, 0.027992412447929382, -0.030093610286712646, 0.05802058428525925, 0.04...
816
816
['Kamil Rocki', 'Tomasz Kornuta', 'Tegan Maharaj']
1610.07675v6
We propose a novel method of regularization for recurrent neural networks called suprisal-driven zoneout. In this method, states zoneout (maintain their previous value rather than updating), when the suprisal (discrepancy between the last state's prediction and target) is small. Thus regularization is adaptive and inpu...
Surprisal-Driven Zoneout
2,016
http://arxiv.org/pdf/1610.07675v6
Title SurprisalDriven Zoneout Summary propose novel method regularization recurrent neural network called suprisaldriven zoneout method state zoneout maintain previous value rather updating suprisal discrepancy last state prediction target small Thus regularization adaptive inputdriven perneuron basis demonstrate effec...
[-0.02612912654876709, 0.04197278246283531, 0.0021873782388865948, 0.008346394635736942, -0.02123868092894554, -0.029539236798882484, -0.0008560279966332018, 0.0529327392578125, -0.08594375848770142, 0.029092682525515556, 0.05646948516368866, -0.0040793693624436855, 0.02608955092728138, 0.03944779932498932, 0.023502251...
817
817
['Barret Zoph', 'Quoc V. Le']
1611.01578v2
Neural networks are powerful and flexible models that work well for many difficult learning tasks in image, speech and natural language understanding. Despite their success, neural networks are still hard to design. In this paper, we use a recurrent network to generate the model descriptions of neural networks and trai...
Neural Architecture Search with Reinforcement Learning
2,016
http://arxiv.org/pdf/1611.01578v2
Title Neural Architecture Search Reinforcement Learning Summary Neural network powerful flexible model work well many difficult learning task image speech natural language understanding Despite success neural network still hard design paper use recurrent network generate model description neural network train RNN reinf...
[0.04315446689724922, 0.041258394718170166, -0.010374544188380241, 0.03191472217440605, -0.0324622243642807, -0.016813019290566444, 0.0007508726557716727, 0.0015609796391800046, -0.047142814844846725, 0.00805315189063549, -0.015890780836343765, -0.03392862156033516, 0.012950747273862362, 0.07452162355184555, 0.01471887...
818
818
['Brian Cheung', 'Eric Weiss', 'Bruno Olshausen']
1611.09430v2
We describe a neural attention model with a learnable retinal sampling lattice. The model is trained on a visual search task requiring the classification of an object embedded in a visual scene amidst background distractors using the smallest number of fixations. We explore the tiling properties that emerge in the mode...
Emergence of foveal image sampling from learning to attend in visual scenes
2,016
http://arxiv.org/pdf/1611.09430v2
Title Emergence foveal image sampling learning attend visual scene Summary describe neural attention model learnable retinal sampling lattice model trained visual search task requiring classification object embedded visual scene amidst background distractors using smallest number fixation explore tiling property emerge...
[0.012428577058017254, -0.03871902450919151, -0.03312397003173828, 0.011463208124041557, 0.024118706583976746, -0.01312296837568283, 0.012522065080702305, -0.016382869333028793, -0.024642430245876312, 0.0025040465407073498, 0.016565145924687386, 0.01324431411921978, 0.0005026690778322518, 0.02636227384209633, 0.0162643...
819
819
['Steven Stenberg Hansen']
1701.03866v1
Credit assignment in traditional recurrent neural networks usually involves back-propagating through a long chain of tied weight matrices. The length of this chain scales linearly with the number of time-steps as the same network is run at each time-step. This creates many problems, such as vanishing gradients, that ha...
Long Timescale Credit Assignment in NeuralNetworks with External Memory
2,017
http://arxiv.org/pdf/1701.03866v1
Title Long Timescale Credit Assignment NeuralNetworks External Memory Summary Credit assignment traditional recurrent neural network usually involves backpropagating long chain tied weight matrix length chain scale linearly number timesteps network run timestep creates many problem vanishing gradient well studied contr...
[0.01334431953728199, 0.025611329823732376, -0.0011260801693424582, 0.027909809723496437, -0.02556299977004528, -0.009520058520138264, 0.04025751352310181, -0.027384348213672638, -0.03948945179581642, 0.00924374908208847, 0.006796651054173708, -0.033932916820049286, 0.014555448666214943, 0.03412463143467903, 0.03704479...
820
820
['Arman Afrasiyabi', 'Ozan Yildiz', 'Baris Nasir', 'Fatos T. Yarman Vural', 'A. Enis Cetin']
1702.02676v1
In recent years, machine learning techniques based on neural networks for mobile computing become increasingly popular. Classical multi-layer neural networks require matrix multiplications at each stage. Multiplication operation is not an energy efficient operation and consequently it drains the battery of the mobile d...
Energy Saving Additive Neural Network
2,017
http://arxiv.org/pdf/1702.02676v1
Title Energy Saving Additive Neural Network Summary recent year machine learning technique based neural network mobile computing become increasingly popular Classical multilayer neural network require matrix multiplication stage Multiplication operation energy efficient operation consequently drain battery mobile devic...
[-0.041559599339962006, 0.0196552537381649, -0.013178504072129726, 0.01698065921664238, 0.020249901339411736, -0.004165116231888533, 0.043071042746305466, -0.03175780177116394, -0.05672718212008476, -0.0013886174419894814, 0.003121745539829135, 0.01433127373456955, 0.022105585783720016, 0.007520826067775488, 0.04837945...
821
821
['Sahil Sharma', 'Aravind S. Lakshminarayanan', 'Balaraman Ravindran']
1702.06054v1
Reinforcement Learning algorithms can learn complex behavioral patterns for sequential decision making tasks wherein an agent interacts with an environment and acquires feedback in the form of rewards sampled from it. Traditionally, such algorithms make decisions, i.e., select actions to execute, at every single time s...
Learning to Repeat: Fine Grained Action Repetition for Deep Reinforcement Learning
2,017
http://arxiv.org/pdf/1702.06054v1
Title Learning Repeat Fine Grained Action Repetition Deep Reinforcement Learning Summary Reinforcement Learning algorithm learn complex behavioral pattern sequential decision making task wherein agent interacts environment acquires feedback form reward sampled Traditionally algorithm make decision ie select action exec...
[-0.007449817378073931, -0.017083393409848213, -0.020155394449830055, -0.014041852205991745, -0.0049669453874230385, 0.01620970480144024, -0.00025756648392416537, -0.018405510112643242, -0.04354967921972275, -0.008511803112924099, -0.0058097923174500465, 0.04968591779470444, -0.00846255011856556, 0.09525875747203827, 0...
822
822
['Amit Sahu']
1702.06186v2
Reason and inference require process as well as memory skills by humans. Neural networks are able to process tasks like image recognition (better than humans) but in memory aspects are still limited (by attention mechanism, size). Recurrent Neural Network (RNN) and it's modified version LSTM are able to solve small mem...
Survey of reasoning using Neural networks
2,017
http://arxiv.org/pdf/1702.06186v2
Title Survey reasoning using Neural network Summary Reason inference require process well memory skill human Neural network able process task like image recognition better human memory aspect still limited attention mechanism size Recurrent Neural Network RNN modified version LSTM able solve small memory context contex...
[0.05290656164288521, 0.045261893421411514, -0.00019537043408490717, 0.06497050076723099, -0.039999376982450485, 0.0035967694129794836, 0.022891681641340256, 0.00871207844465971, 0.014085907489061356, -0.042083173990249634, 0.01758415438234806, 0.008582827635109425, 0.004018485080450773, 0.018058646470308304, 0.0189506...
823
823
['Yan Duan', 'Marcin Andrychowicz', 'Bradly C. Stadie', 'Jonathan Ho', 'Jonas Schneider', 'Ilya Sutskever', 'Pieter Abbeel', 'Wojciech Zaremba']
1703.07326v3
Imitation learning has been commonly applied to solve different tasks in isolation. This usually requires either careful feature engineering, or a significant number of samples. This is far from what we desire: ideally, robots should be able to learn from very few demonstrations of any given task, and instantly general...
One-Shot Imitation Learning
2,017
http://arxiv.org/pdf/1703.07326v3
Title OneShot Imitation Learning Summary Imitation learning commonly applied solve different task isolation usually requires either careful feature engineering significant number sample far desire ideally robot able learn demonstration given task instantly generalize new situation task without requiring taskspecific en...
[-0.02619013376533985, 0.003925428260117769, -0.004163840785622597, -0.022319246083498, -0.05592583492398262, -0.021797653287649155, 0.024787820875644684, 0.01689254678785801, -0.011839345097541809, -0.00804369617253542, -0.004566506016999483, -0.030544642359018326, -0.024636536836624146, 0.02313395030796528, 0.0157427...
824
824
['Shumeet Baluja']
1703.07394v1
In all but the most trivial optimization problems, the structure of the solutions exhibit complex interdependencies between the input parameters. Decades of research with stochastic search techniques has shown the benefit of explicitly modeling the interactions between sets of parameters and the overall quality of the ...
Deep Learning for Explicitly Modeling Optimization Landscapes
2,017
http://arxiv.org/pdf/1703.07394v1
Title Deep Learning Explicitly Modeling Optimization Landscapes Summary trivial optimization problem structure solution exhibit complex interdependency input parameter Decades research stochastic search technique shown benefit explicitly modeling interaction set parameter overall quality solution discovered demonstrate...
[-0.010769948363304138, 0.0350692979991436, -0.03418678790330887, 0.027825459837913513, -0.011421464383602142, -0.0665297731757164, 0.013387186452746391, 0.00261049740947783, -0.08670773357152939, 0.004274922888725996, 0.008350112475454807, 0.007464745081961155, -0.007705566473305225, 0.07621415704488754, 0.01865247450...
825
825
['Carlos Florensa', 'Yan Duan', 'Pieter Abbeel']
1704.03012v1
Deep reinforcement learning has achieved many impressive results in recent years. However, tasks with sparse rewards or long horizons continue to pose significant challenges. To tackle these important problems, we propose a general framework that first learns useful skills in a pre-training environment, and then levera...
Stochastic Neural Networks for Hierarchical Reinforcement Learning
2,017
http://arxiv.org/pdf/1704.03012v1
Title Stochastic Neural Networks Hierarchical Reinforcement Learning Summary Deep reinforcement learning achieved many impressive result recent year However task sparse reward long horizon continue pose significant challenge tackle important problem propose general framework first learns useful skill pretraining enviro...
[-0.0013203145936131477, 0.04116049036383629, -0.004599382169544697, -0.008501452393829823, 0.022824062034487724, -0.02133704349398613, 0.023822534829378128, -0.008480294607579708, -0.051138367503881454, 0.02041446790099144, -0.06494960933923721, 0.0006839332054369152, 0.0030479920096695423, 0.07526356726884842, 0.0206...
826
826
['Daniel Hein', 'Steffen Udluft', 'Michel Tokic', 'Alexander Hentschel', 'Thomas A. Runkler', 'Volkmar Sterzing']
1705.07262v2
The Particle Swarm Optimization Policy (PSO-P) has been recently introduced and proven to produce remarkable results on interacting with academic reinforcement learning benchmarks in an off-policy, batch-based setting. To further investigate the properties and feasibility on real-world applications, this paper investig...
Batch Reinforcement Learning on the Industrial Benchmark: First Experiences
2,017
http://arxiv.org/pdf/1705.07262v2
Title Batch Reinforcement Learning Industrial Benchmark First Experiences Summary Particle Swarm Optimization Policy PSOP recently introduced proven produce remarkable result interacting academic reinforcement learning benchmark offpolicy batchbased setting investigate property feasibility realworld application paper i...
[-0.04859529808163643, 0.009126704186201096, 0.004413751419633627, -0.036701805889606476, -0.009103127755224705, -0.053609319031238556, 0.0009935915004462004, 0.02723786048591137, -0.0376310721039772, 0.0017250364180654287, 0.024324486032128334, 0.02293894998729229, 0.003281214740127325, 0.029910871759057045, 0.0144245...
827
827
['Tim Rocktäschel', 'Sebastian Riedel']
1705.11040v2
We introduce neural networks for end-to-end differentiable proving of queries to knowledge bases by operating on dense vector representations of symbols. These neural networks are constructed recursively by taking inspiration from the backward chaining algorithm as used in Prolog. Specifically, we replace symbolic unif...
End-to-End Differentiable Proving
2,017
http://arxiv.org/pdf/1705.11040v2
Title EndtoEnd Differentiable Proving Summary introduce neural network endtoend differentiable proving query knowledge base operating dense vector representation symbol neural network constructed recursively taking inspiration backward chaining algorithm used Prolog Specifically replace symbolic unification differentia...
[0.00876219104975462, 0.0424051396548748, -0.019259411841630936, 0.040070559829473495, -0.024717167019844055, -0.019320471212267876, 0.026828352361917496, 0.020608365535736084, -0.009044548496603966, 0.003754157805815339, 0.0062873768620193005, 0.061944689601659775, -0.0130495959892869, 0.041643187403678894, 0.02143714...
828
828
['Ryan Lowe', 'Yi Wu', 'Aviv Tamar', 'Jean Harb', 'Pieter Abbeel', 'Igor Mordatch']
1706.02275v3
We explore deep reinforcement learning methods for multi-agent domains. We begin by analyzing the difficulty of traditional algorithms in the multi-agent case: Q-learning is challenged by an inherent non-stationarity of the environment, while policy gradient suffers from a variance that increases as the number of agent...
Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
2,017
http://arxiv.org/pdf/1706.02275v3
Title MultiAgent ActorCritic Mixed CooperativeCompetitive Environments Summary explore deep reinforcement learning method multiagent domain begin analyzing difficulty traditional algorithm multiagent case Qlearning challenged inherent nonstationarity environment policy gradient suffers variance increase number agent gr...
[0.014611853286623955, 0.02163694240152836, -0.022250503301620483, -0.018286388367414474, -0.019145065918564796, -0.006084064487367868, 0.04140058159828186, -0.05955472216010094, -0.03481706976890564, 0.007267179433256388, -0.03561222925782204, -0.021177997812628746, -0.03134753555059433, 0.023660384118556976, -0.00548...
829
829
['Joan Serrà', 'Alexandros Karatzoglou']
1706.03993v1
Recommendation algorithms that incorporate techniques from deep learning are becoming increasingly popular. Due to the structure of the data coming from recommendation domains (i.e., one-hot-encoded vectors of item preferences), these algorithms tend to have large input and output dimensionalities that dominate their o...
Getting deep recommenders fit: Bloom embeddings for sparse binary input/output networks
2,017
http://arxiv.org/pdf/1706.03993v1
Title Getting deep recommenders fit Bloom embeddings sparse binary inputoutput network Summary Recommendation algorithm incorporate technique deep learning becoming increasingly popular Due structure data coming recommendation domain ie onehotencoded vector item preference algorithm tend large input output dimensionali...
[0.012883339077234268, -0.009338920004665852, -0.01198730431497097, 0.009027131833136082, -0.002386106178164482, -0.04708776995539665, 0.05819787085056305, 0.016367528587579727, 0.010520237497985363, -0.018471479415893555, 0.008583184331655502, -0.014873209409415722, -0.009784399531781673, 0.12243649363517761, 0.010296...
830
830
['Jinzhuo Wang', 'Wenmin Wang', 'Ronggang Wang', 'Wen Gao']
1706.04052v1
Monte Carlo tree search (MCTS) is extremely popular in computer Go which determines each action by enormous simulations in a broad and deep search tree. However, human experts select most actions by pattern analysis and careful evaluation rather than brute search of millions of future nteractions. In this paper, we pro...
Beyond Monte Carlo Tree Search: Playing Go with Deep Alternative Neural Network and Long-Term Evaluation
2,017
http://arxiv.org/pdf/1706.04052v1
Title Beyond Monte Carlo Tree Search Playing Go Deep Alternative Neural Network LongTerm Evaluation Summary Monte Carlo tree search MCTS extremely popular computer Go determines action enormous simulation broad deep search tree However human expert select action pattern analysis careful evaluation rather brute search m...
[-0.018440835177898407, 0.00823884829878807, -0.025559427216649055, 0.01030297763645649, 0.009421521797776222, -0.06514395773410797, -0.020029939711093903, 0.028141772374510765, -0.04764072597026825, 0.010844013653695583, -0.01397472620010376, 0.01675931178033352, -0.020373951643705368, 0.07995585352182388, 0.044008787...
831
831
['Marcin Andrychowicz', 'Filip Wolski', 'Alex Ray', 'Jonas Schneider', 'Rachel Fong', 'Peter Welinder', 'Bob McGrew', 'Josh Tobin', 'Pieter Abbeel', 'Wojciech Zaremba']
1707.01495v3
Dealing with sparse rewards is one of the biggest challenges in Reinforcement Learning (RL). We present a novel technique called Hindsight Experience Replay which allows sample-efficient learning from rewards which are sparse and binary and therefore avoid the need for complicated reward engineering. It can be combined...
Hindsight Experience Replay
2,017
http://arxiv.org/pdf/1707.01495v3
Title Hindsight Experience Replay Summary Dealing sparse reward one biggest challenge Reinforcement Learning RL present novel technique called Hindsight Experience Replay allows sampleefficient learning reward sparse binary therefore avoid need complicated reward engineering combined arbitrary offpolicy RL algorithm ma...
[0.01251895260065794, -0.006217009387910366, 0.008013303391635418, -0.048185672610998154, -0.024119410663843155, -0.022304944694042206, 0.00254358584061265, 0.012106338515877724, -0.0323614664375782, -0.0015307181747630239, -0.002653340343385935, 0.03844265267252922, -0.033091168850660324, 0.04591533914208412, -0.02303...
832
832
['William Saunders', 'Girish Sastry', 'Andreas Stuhlmueller', 'Owain Evans']
1707.05173v1
AI systems are increasingly applied to complex tasks that involve interaction with humans. During training, such systems are potentially dangerous, as they haven't yet learned to avoid actions that could cause serious harm. How can an AI system explore and learn without making a single mistake that harms humans or othe...
Trial without Error: Towards Safe Reinforcement Learning via Human Intervention
2,017
http://arxiv.org/pdf/1707.05173v1
Title Trial without Error Towards Safe Reinforcement Learning via Human Intervention Summary AI system increasingly applied complex task involve interaction human training system potentially dangerous havent yet learned avoid action could cause serious harm AI system explore learn without making single mistake harm hum...
[0.016132855787873268, 0.04305306077003479, -0.005523810163140297, -0.06343850493431091, 0.003946242388337851, 0.03117883950471878, -1.495579181209905e-05, -0.007241299841552973, -1.276238435821142e-06, 0.02175363525748253, 0.02454415149986744, 0.05398100987076759, -0.042051032185554504, 0.0768401250243187, 0.029091153...
833
833
['Carlos Florensa', 'David Held', 'Markus Wulfmeier', 'Michael Zhang', 'Pieter Abbeel']
1707.05300v2
Many relevant tasks require an agent to reach a certain state, or to manipulate objects into a desired configuration. For example, we might want a robot to align and assemble a gear onto an axle or insert and turn a key in a lock. These goal-oriented tasks present a considerable challenge for reinforcement learning, si...
Reverse Curriculum Generation for Reinforcement Learning
2,017
http://arxiv.org/pdf/1707.05300v2
Title Reverse Curriculum Generation Reinforcement Learning Summary Many relevant task require agent reach certain state manipulate object desired configuration example might want robot align assemble gear onto axle insert turn key lock goaloriented task present considerable challenge reinforcement learning since natura...
[0.02395312860608101, -0.03560492396354675, -0.022230951115489006, -0.027542339637875557, -0.009401685558259487, -0.01566183567047119, -0.00923572201281786, -0.021622776985168457, -0.03852638974785805, -0.0009287800639867783, -0.02003348246216774, 0.06214943900704384, -0.010654082521796227, 0.04130471497774124, -0.0070...
834
834
['S. Hossein Hosseini', 'Afshin Ebrahimi']
1707.06992v2
A population-based optimization algorithm was designed, inspired by two main thinking modes in philosophy, both based on dialectic concept and thesis-antithesis paradigm. They impose two different kinds of dialectics. Idealistic and materialistic antitheses are formulated as optimization models. Based on the models, th...
Ideological Sublations: Resolution of Dialectic in Population-based Optimization
2,017
http://arxiv.org/pdf/1707.06992v2
Title Ideological Sublations Resolution Dialectic Populationbased Optimization Summary populationbased optimization algorithm designed inspired two main thinking mode philosophy based dialectic concept thesisantithesis paradigm impose two different kind dialectic Idealistic materialistic antithesis formulated optimizat...
[0.008950497023761272, 0.06336093693971634, -0.013111977837979794, -0.01586339622735977, -0.02789732627570629, -0.04434084892272949, 0.017018530517816544, 0.016037438064813614, -0.07308778911828995, -0.0345175601541996, 0.050040144473314285, 0.0054401527158916, 0.007886533625423908, 0.009143506176769733, -0.04539664462...
835
835
['Sujith Ravi']
1708.00630v2
Deep neural networks have become ubiquitous for applications related to visual recognition and language understanding tasks. However, it is often prohibitive to use typical neural networks on devices like mobile phones or smart watches since the model sizes are huge and cannot fit in the limited memory available on suc...
ProjectionNet: Learning Efficient On-Device Deep Networks Using Neural Projections
2,017
http://arxiv.org/pdf/1708.00630v2
Title ProjectionNet Learning Efficient OnDevice Deep Networks Using Neural Projections Summary Deep neural network become ubiquitous application related visual recognition language understanding task However often prohibitive use typical neural network device like mobile phone smart watch since model size huge cannot f...
[0.009888674132525921, 0.028102440759539604, -0.004740584641695023, 0.06274411827325821, -0.011353502050042152, -0.021599646657705307, 0.03855704516172409, -0.020037738606333733, -0.060172080993652344, 0.003622456919401884, -0.037688616663217545, 0.002514038933441043, 0.041991688311100006, 0.011390387080609798, -0.0009...
836
836
['Zhen Li', 'Zuoqiang Shi']
1708.06257v2
Based on a natural connection between ResNet and transport equation or its characteristic equation, we propose a continuous flow model for both ResNet and plain net. Through this continuous model, a ResNet can be explicitly constructed as a refinement of a plain net. The flow model provides an alternative perspective t...
A Flow Model of Neural Networks
2,017
http://arxiv.org/pdf/1708.06257v2
Title Flow Model Neural Networks Summary Based natural connection ResNet transport equation characteristic equation propose continuous flow model ResNet plain net continuous model ResNet explicitly constructed refinement plain net flow model provides alternative perspective understand phenomenon deep neural network nec...
[-0.05974646657705307, -0.013680173084139824, -0.041853632777929306, 0.01939602755010128, -0.024163352325558662, -0.051294393837451935, 0.005934596993029118, -0.03236718475818634, -0.03148850426077843, 0.0065529001876711845, 0.05397547036409378, 0.013033125549554825, 0.007781938184052706, 0.05482274293899536, 0.0474607...
837
837
['Nikhita Vedula', 'Wei Sun', 'Hyunhwan Lee', 'Harsh Gupta', 'Mitsunori Ogihara', 'Joseph Johnson', 'Gang Ren', 'Srinivasan Parthasarathy']
1709.03946v1
The rapid advances in e-commerce and Web 2.0 technologies have greatly increased the impact of commercial advertisements on the general public. As a key enabling technology, a multitude of recommender systems exists which analyzes user features and browsing patterns to recommend appealing advertisements to users. In th...
Multimodal Content Analysis for Effective Advertisements on YouTube
2,017
http://arxiv.org/pdf/1709.03946v1
Title Multimodal Content Analysis Effective Advertisements YouTube Summary rapid advance ecommerce Web 20 technology greatly increased impact commercial advertisement general public key enabling technology multitude recommender system exists analyzes user feature browsing pattern recommend appealing advertisement user ...
[0.05325677990913391, 0.01905878633260727, -0.04145479202270508, -0.01872815005481243, 0.0188889279961586, -0.016468115150928497, 0.04491739347577095, -0.013797054998576641, -0.022574132308363914, -0.10200163722038269, -0.054233718663454056, -0.02793099358677864, -0.001039214781485498, 0.10229802876710892, 0.0115968361...
838
838
['Ashvin Nair', 'Bob McGrew', 'Marcin Andrychowicz', 'Wojciech Zaremba', 'Pieter Abbeel']
1709.10089v2
Exploration in environments with sparse rewards has been a persistent problem in reinforcement learning (RL). Many tasks are natural to specify with a sparse reward, and manually shaping a reward function can result in suboptimal performance. However, finding a non-zero reward is exponentially more difficult with incre...
Overcoming Exploration in Reinforcement Learning with Demonstrations
2,017
http://arxiv.org/pdf/1709.10089v2
Title Overcoming Exploration Reinforcement Learning Demonstrations Summary Exploration environment sparse reward persistent problem reinforcement learning RL Many task natural specify sparse reward manually shaping reward function result suboptimal performance However finding nonzero reward exponentially difficult incr...
[-0.000670247187372297, 0.007406280376017094, 0.008679671213030815, -0.04938372224569321, -0.01916898600757122, -0.008341330103576183, -0.0046310280449688435, -0.032446566969156265, -0.055457938462495804, 0.03828710317611694, -0.02534225583076477, 0.06884787976741791, -0.0331156887114048, 0.07698538899421692, -0.000784...
839
839
['Chaitanya Ahuja', 'Louis-Philippe Morency']
1710.02254v2
Recurrent neural networks have shown remarkable success in modeling sequences. However low resource situations still adversely affect the generalizability of these models. We introduce a new family of models, called Lattice Recurrent Units (LRU), to address the challenge of learning deep multi-layer recurrent models wi...
Lattice Recurrent Unit: Improving Convergence and Statistical Efficiency for Sequence Modeling
2,017
http://arxiv.org/pdf/1710.02254v2
Title Lattice Recurrent Unit Improving Convergence Statistical Efficiency Sequence Modeling Summary Recurrent neural network shown remarkable success modeling sequence However low resource situation still adversely affect generalizability model introduce new family model called Lattice Recurrent Units LRU address chall...
[-0.02703038603067398, 0.056324027478694916, -0.019291691482067108, 0.06274662911891937, -0.03949630632996559, -0.0014617762062698603, 0.028200378641486168, -0.00469781132414937, -0.07082455605268478, -0.008138565346598625, 0.022258136421442032, -0.07556074857711792, 0.033635951578617096, 0.027700180187821388, 0.006997...
840
840
['Matthew Riemer', 'Tim Klinger', 'Michele Franceschini', 'Djallel Bouneffouf']
1711.06761v3
Given the recent success of Deep Learning applied to a variety of single tasks, it is natural to consider more human-realistic settings. Perhaps the most difficult of these settings is that of continual lifelong learning, where the model must learn online over a continuous stream of non-stationary data. A continual lif...
Scalable Recollections for Continual Lifelong Learning
2,017
http://arxiv.org/pdf/1711.06761v3
Title Scalable Recollections Continual Lifelong Learning Summary Given recent success Deep Learning applied variety single task natural consider humanrealistic setting Perhaps difficult setting continual lifelong learning model must learn online continuous stream nonstationary data continual lifelong learning system mu...
[0.02174822986125946, 0.05593891069293022, -0.015595111064612865, 0.010836074128746986, -0.020778335630893707, 0.005336886737495661, 0.04968583956360817, -0.042867522686719894, -0.022607678547501564, -0.007372090592980385, -0.0037958917673677206, -0.03462539613246918, -0.06478039175271988, 0.09908340126276016, -0.02315...
841
841
['Davide Bacciu']
1711.07784v1
The paper introduces the Hidden Tree Markov Network (HTN), a neuro-probabilistic hybrid fusing the representation power of generative models for trees with the incremental and discriminative learning capabilities of neural networks. We put forward a modular architecture in which multiple generative models of limited co...
Hidden Tree Markov Networks: Deep and Wide Learning for Structured Data
2,017
http://arxiv.org/pdf/1711.07784v1
Title Hidden Tree Markov Networks Deep Wide Learning Structured Data Summary paper introduces Hidden Tree Markov Network HTN neuroprobabilistic hybrid fusing representation power generative model tree incremental discriminative learning capability neural network put forward modular architecture multiple generative mode...
[0.011593655683100224, 0.006323219742625952, 0.025108326226472855, 0.05469093099236488, -0.0006732799229212105, 0.027713587507605553, -0.053613532334566116, 0.0014808099949732423, -0.013125276193022728, -0.07022825628519058, -0.0016904890071600676, -0.006906906142830849, 0.01242598332464695, 0.08071021735668182, -0.015...
842
842
['Andrew Levy', 'Robert Platt', 'Kate Saenko']
1712.00948v3
The ability to learn at different resolutions in time may help overcome one of the main challenges in deep reinforcement learning -- sample efficiency. Hierarchical agents that operate at different levels of temporal abstraction can learn tasks more quickly because they can divide the work of learning behaviors among m...
Hierarchical Actor-Critic
2,017
http://arxiv.org/pdf/1712.00948v3
Title Hierarchical ActorCritic Summary ability learn different resolution time may help overcome one main challenge deep reinforcement learning sample efficiency Hierarchical agent operate different level temporal abstraction learn task quickly divide work learning behavior among multiple policy also explore environmen...
[0.010711533948779106, 0.057223305106163025, -0.029990557581186295, -0.02625054307281971, 0.0012431297218427062, 0.007482779212296009, 0.04047413542866707, 0.01626020297408104, -0.007949854247272015, -0.00869183149188757, -0.0275528896600008, -0.01351903285831213, -0.04517769441008568, 0.08610741794109344, -0.002952864...
843
843
['Freek Stulp', 'Pierre-Yves Oudeyer']
1712.05249v1
To harness the complexity of their high-dimensional bodies during sensorimotor development, infants are guided by patterns of freezing and freeing of degrees of freedom. For instance, when learning to reach, infants free the degrees of freedom in their arm proximodistally, i.e. from joints that are closer to the body t...
Proximodistal Exploration in Motor Learning as an Emergent Property of Optimization
2,017
http://arxiv.org/pdf/1712.05249v1
Title Proximodistal Exploration Motor Learning Emergent Property Optimization Summary harness complexity highdimensional body sensorimotor development infant guided pattern freezing freeing degree freedom instance learning reach infant free degree freedom arm proximodistally ie joint closer body distant formulate study...
[-0.007591096218675375, -0.030919570475816727, -0.04300932586193085, -0.013939157128334045, -0.01064420398324728, 0.022560911253094673, 0.019957570359110832, -0.005076027940958738, -0.029832862317562103, 0.026859911158680916, -0.024408021941781044, -0.03122008964419365, 0.02185315266251564, 0.028434840962290764, 0.0178...
844
844
['M. J. Gagen']
1712.09014v1
The hard problem in artificial intelligence asks how the shuffling of syntactical symbols in a program can lead to systems which experience semantics and qualia. We address this question in three stages. First, we introduce a new class of human semantic symbols which appears when unexpected and drastic environmental ch...
Null Dynamical State Models of Human Cognitive Dysfunction
2,017
http://arxiv.org/pdf/1712.09014v1
Title Null Dynamical State Models Human Cognitive Dysfunction Summary hard problem artificial intelligence asks shuffling syntactical symbol program lead system experience semantics qualia address question three stage First introduce new class human semantic symbol appears unexpected drastic environmental change cause ...
[0.02669551596045494, 0.01971989870071411, -0.04531700164079666, -0.03280062973499298, -0.0007659009424969554, 0.03856762871146202, -0.01750306226313114, 0.026557743549346924, 0.042881399393081665, -0.0329393595457077, 0.03975263983011246, 0.027196673676371574, 0.006306413095444441, 0.057519715279340744, 0.018768649548...
845
845
['Haik Manukian', 'Fabio L. Traversa', 'Massimiliano Di Ventra']
1801.00512v2
Restricted Boltzmann machines (RBMs) and their extensions, called 'deep-belief networks', are powerful neural networks that have found applications in the fields of machine learning and big data. The standard way to training these models resorts to an iterative unsupervised procedure based on Gibbs sampling, called 'co...
Accelerating Deep Learning with Memcomputing
2,018
http://arxiv.org/pdf/1801.00512v2
Title Accelerating Deep Learning Memcomputing Summary Restricted Boltzmann machine RBMs extension called deepbelief network powerful neural network found application field machine learning big data standard way training model resort iterative unsupervised procedure based Gibbs sampling called contrastive divergence CD ...
[-0.021321767941117287, 0.008590729907155037, -0.04020331799983978, 0.0603061243891716, 0.013501854613423347, 0.004984626546502113, 0.006701979320496321, -0.006113486364483833, -0.0337209552526474, 0.019319409504532814, -0.03536015748977661, 0.015597990714013577, -0.016139866784214973, 0.03549143299460411, 0.0296674575...
846
846
['Yu Shi', 'Fangqiu Han', 'Xinran He', 'Carl Yang', 'Jie Luo', 'Jiawei Han']
1801.06597v1
Multi-view networks are ubiquitous in real-world applications. In order to extract knowledge or business value, it is of interest to transform such networks into representations that are easily machine-actionable. Meanwhile, network embedding has emerged as an effective approach to generate distributed network represen...
mvn2vec: Preservation and Collaboration in Multi-View Network Embedding
2,018
http://arxiv.org/pdf/1801.06597v1
Title mvn2vec Preservation Collaboration MultiView Network Embedding Summary Multiview network ubiquitous realworld application order extract knowledge business value interest transform network representation easily machineactionable Meanwhile network embedding emerged effective approach generate distributed network re...
[-0.0012199279153719544, 0.03439829498529434, -0.014865078963339329, 0.005650732666254044, 0.013211688958108425, -0.023278461769223213, -0.00024796868092380464, -0.02037089131772518, 0.020105179399251938, -0.020993001759052277, 0.036786191165447235, 0.02955283224582672, -0.0013284682063385844, 0.06180468946695328, 0.03...
847
847
['Patrick Schwab', 'Walter Karlen']
1802.02195v1
Several methods have recently been proposed to detect salient input features for outputs of neural networks. Those methods offer a qualitative glimpse at feature importance, but they fall short of providing quantifiable attributions that can be compared across decisions and measures of the expected quality of their exp...
Granger-causal Attentive Mixtures of Experts
2,018
http://arxiv.org/pdf/1802.02195v1
Title Grangercausal Attentive Mixtures Experts Summary Several method recently proposed detect salient input feature output neural network method offer qualitative glimpse feature importance fall short providing quantifiable attribution compared across decision measure expected quality explanation address shortcoming p...
[0.00200763251632452, 0.010885958559811115, -0.006908542010933161, 0.019717030227184296, 0.01309158094227314, 0.019467514008283615, 0.0177998598664999, -0.010452842339873314, -0.03912140801548958, 0.004165454301983118, -0.014568503014743328, 0.013057363219559193, 0.04707609862089157, 0.036716509610414505, 0.01283021923...
848
848
['Adam Liška', 'Germán Kruszewski', 'Marco Baroni']
1802.06467v1
Neural networks are very powerful learning systems, but they do not readily generalize from one task to the other. This is partly due to the fact that they do not learn in a compositional way, that is, by discovering skills that are shared by different tasks, and recombining them to solve new problems. In this paper, w...
Memorize or generalize? Searching for a compositional RNN in a haystack
2,018
http://arxiv.org/pdf/1802.06467v1
Title Memorize generalize Searching compositional RNN haystack Summary Neural network powerful learning system readily generalize one task partly due fact learn compositional way discovering skill shared different task recombining solve new problem paper explore compositional generalization capability recurrent neural ...
[0.029544081538915634, 0.060220688581466675, -0.0016016733134165406, -0.006507180165499449, -0.018431834876537323, -0.021829428151249886, 0.0024944611359387636, -0.01524284016340971, -0.07089319080114365, 0.004815997555851936, -0.011547374539077282, -0.035060349851846695, 0.006641760002821684, 0.00652617821469903, 0.02...
849
849
['Christos Kaplanis', 'Murray Shanahan', 'Claudia Clopath']
1802.07239v1
Unlike humans, who are capable of continual learning over their lifetimes, artificial neural networks have long been known to suffer from a phenomenon known as catastrophic forgetting, whereby new learning can lead to abrupt erasure of previously acquired knowledge. Whereas in a neural network the parameters are typica...
Continual Reinforcement Learning with Complex Synapses
2,018
http://arxiv.org/pdf/1802.07239v1
Title Continual Reinforcement Learning Complex Synapses Summary Unlike human capable continual learning lifetime artificial neural network long known suffer phenomenon known catastrophic forgetting whereby new learning lead abrupt erasure previously acquired knowledge Whereas neural network parameter typically modelled...
[-0.01147836446762085, 0.009462928399443626, -0.021536637097597122, -0.05127372220158577, 0.006826843600720167, 0.022359762340784073, -0.010852308012545109, -0.03854924067854881, -0.016124475747346878, 0.0515826940536499, -0.006582262925803661, -0.002793267834931612, -0.019202113151550293, 0.04928802698850632, 0.002875...
850
850
['Abhishek Gupta', 'Russell Mendonca', 'YuXuan Liu', 'Pieter Abbeel', 'Sergey Levine']
1802.07245v1
Exploration is a fundamental challenge in reinforcement learning (RL). Many of the current exploration methods for deep RL use task-agnostic objectives, such as information gain or bonuses based on state visitation. However, many practical applications of RL involve learning more than a single task, and prior tasks can...
Meta-Reinforcement Learning of Structured Exploration Strategies
2,018
http://arxiv.org/pdf/1802.07245v1
Title MetaReinforcement Learning Structured Exploration Strategies Summary Exploration fundamental challenge reinforcement learning RL Many current exploration method deep RL use taskagnostic objective information gain bonus based state visitation However many practical application RL involve learning single task prior...
[0.022845996543765068, -0.010327394120395184, 0.00637950049713254, -0.044209763407707214, 0.012235644273459911, -0.014252508990466595, 0.008930147625505924, -0.013155504129827023, -0.06443705409765244, -0.008876400999724865, -0.02665579505264759, 0.05503110960125923, -0.04755299538373947, 0.0789950042963028, -0.0079069...
851
851
['Matthew Streeter']
1802.07697v1
We present an approximation algorithm that takes a pool of pre-trained models as input and produces from it a cascaded model with similar accuracy but lower average-case cost. Applied to state-of-the-art ImageNet classification models, this yields up to a 2x reduction in floating point multiplications, and up to a 6x r...
Approximation Algorithms for Cascading Prediction Models
2,018
http://arxiv.org/pdf/1802.07697v1
Title Approximation Algorithms Cascading Prediction Models Summary present approximation algorithm take pool pretrained model input produce cascaded model similar accuracy lower averagecase cost Applied stateoftheart ImageNet classification model yield 2x reduction floating point multiplication 6x reduction averagecase...
[-0.01582586206495762, 0.03970373049378395, -0.02020801231265068, 0.02555765211582184, -0.009261867962777615, 0.013759392313659191, 0.0602111853659153, 0.026877330616116524, -0.01674836501479149, 0.02368089184165001, 0.010600359179079533, -0.036255110055208206, 0.01976088061928749, -0.00016322686860803515, -0.037982922...
852
852
['Wlodzislaw Duch']
1802.08478v1
Neural networks are commonly regarded as black boxes performing incomprehensible functions. For classification problems networks provide maps from high dimensional feature space to K-dimensional image space. Images of training vector are projected on polygon vertices, providing visualization of network function. Such v...
Coloring black boxes: visualization of neural network decisions
2,018
http://arxiv.org/pdf/1802.08478v1
Title Coloring black box visualization neural network decision Summary Neural network commonly regarded black box performing incomprehensible function classification problem network provide map high dimensional feature space Kdimensional image space Images training vector projected polygon vertex providing visualizatio...
[-0.027500305324792862, 0.00057623052271083, -0.024883462116122246, 0.01702534779906273, -0.04228878766298294, -0.013837902806699276, 0.06630858778953552, 0.03646942973136902, -0.016635505482554436, 0.013931164517998695, 0.020103393122553825, 0.035459134727716446, 0.01946960762143135, 0.08091599494218826, 0.03291395679...
853
853
['Sjoerd van Steenkiste', 'Michael Chang', 'Klaus Greff', 'Jürgen Schmidhuber']
1802.10353v1
Common-sense physical reasoning is an essential ingredient for any intelligent agent operating in the real-world. For example, it can be used to simulate the environment, or to infer the state of parts of the world that are currently unobserved. In order to match real-world conditions this causal knowledge must be lear...
Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions
2,018
http://arxiv.org/pdf/1802.10353v1
Title Relational Neural Expectation Maximization Unsupervised Discovery Objects Interactions Summary Commonsense physical reasoning essential ingredient intelligent agent operating realworld example used simulate environment infer state part world currently unobserved order match realworld condition causal knowledge mu...
[0.012887672521173954, 0.04423285648226738, 0.008632563054561615, 0.07505037635564804, -0.009485420770943165, -0.02576008066534996, 0.017814286053180695, -0.03150689974427223, 0.03437905013561249, -0.003737053833901882, 0.011476302519440651, 0.010449493303894997, 0.0022367045748978853, 0.07655835151672363, 0.0180681329...
854
854
['Henrique X. Goulart', 'Guilherme A. Wachs-Lopes']
1802.10393v1
In events that are composed by many activities, there is a problem that involves retrieve and management the information of visitors that are visiting the activities. This management is crucial to find some activities that are drawing attention of visitors; identify an ideal positioning for activities; which path is mo...
A Bayesian Model for Activities Recommendation and Event Structure Optimization Using Visitors Tracking
2,018
http://arxiv.org/pdf/1802.10393v1
Title Bayesian Model Activities Recommendation Event Structure Optimization Using Visitors Tracking Summary event composed many activity problem involves retrieve management information visitor visiting activity management crucial find activity drawing attention visitor identify ideal positioning activity path frequent...
[-0.02488214522600174, -0.020626138895750046, -0.025866858661174774, 0.002105511724948883, -0.015047399327158928, -0.04339215159416199, -0.009036407805979252, -0.015825681388378143, 0.01833384670317173, -0.026449795812368393, -0.002201515482738614, -0.00040884671034291387, 0.018582487478852272, 0.06292422115802765, -0....
855
855
['Jonathan Frankle', 'Michael Carbin']
1803.03635v1
Recent work on neural network pruning indicates that, at training time, neural networks need to be significantly larger in size than is necessary to represent the eventual functions that they learn. This paper articulates a new hypothesis to explain this phenomenon. This conjecture, which we term the "lottery ticket hy...
The Lottery Ticket Hypothesis: Training Pruned Neural Networks
2,018
http://arxiv.org/pdf/1803.03635v1
Title Lottery Ticket Hypothesis Training Pruned Neural Networks Summary Recent work neural network pruning indicates training time neural network need significantly larger size necessary represent eventual function learn paper articulates new hypothesis explain phenomenon conjecture term lottery ticket hypothesis propo...
[0.004814235493540764, 0.05187557265162468, -0.036668747663497925, 0.024571796879172325, -0.029595941305160522, -0.01452797744423151, 0.025005890056490898, 0.022376859560608864, -0.06452710926532745, 0.029688477516174316, 0.007371921092271805, -0.062144190073013306, 0.014480029232800007, 0.0454886369407177, 0.021643474...
856
856
['Christopher Kim', 'Carson Chow']
1803.06622v1
Spiking activity of neurons engaged in learning and performing a task show complex spatiotemporal dynamics. While the output of recurrent network models can learn to perform various tasks, the possible range of recurrent dynamics that emerge after learning remains unknown. Here we show that modifying the recurrent conn...
Learning recurrent dynamics in spiking networks
2,018
http://arxiv.org/pdf/1803.06622v1
Title Learning recurrent dynamic spiking network Summary Spiking activity neuron engaged learning performing task show complex spatiotemporal dynamic output recurrent network model learn perform various task possible range recurrent dynamic emerge learning remains unknown show modifying recurrent connectivity recursive...
[-0.03337254002690315, -0.05719134956598282, -0.017811812460422516, -0.0037006523925811052, 0.015769802033901215, 0.019087083637714386, -0.02505081333220005, -0.021392086520791054, -0.009612438268959522, 0.004935021046549082, -0.009756585583090782, 0.0027564247138798237, 0.02311229519546032, 0.09632956981658936, 0.0255...
857
857
['A. N. Gorban', 'A. Y. Zinovyev']
0809.0490v2
In many physical, statistical, biological and other investigations it is desirable to approximate a system of points by objects of lower dimension and/or complexity. For this purpose, Karl Pearson invented principal component analysis in 1901 and found 'lines and planes of closest fit to system of points'. The famous k...
Principal Graphs and Manifolds
2,008
http://arxiv.org/pdf/0809.0490v2
Title Principal Graphs Manifolds Summary many physical statistical biological investigation desirable approximate system point object lower dimension andor complexity purpose Karl Pearson invented principal component analysis 1901 found line plane closest fit system point famous kmeans algorithm solves approximation pr...
[-0.05076771602034569, -0.029868997633457184, -0.0129631869494915, 0.026080191135406494, -0.014954281970858574, -0.023005656898021698, 0.009431012906134129, -0.003656127257272601, 0.014218808151781559, 0.019609374925494194, 0.0367770753800869, 0.01792166940867901, 0.013533476740121841, 0.017544740810990334, 0.051993824...
858
858
['Xanadu Halkias', 'Sebastien Paris', 'Herve Glotin']
1301.3533v2
Deep Belief Networks (DBN) have been successfully applied on popular machine learning tasks. Specifically, when applied on hand-written digit recognition, DBNs have achieved approximate accuracy rates of 98.8%. In an effort to optimize the data representation achieved by the DBN and maximize their descriptive power, re...
Sparse Penalty in Deep Belief Networks: Using the Mixed Norm Constraint
2,013
http://arxiv.org/pdf/1301.3533v2
Title Sparse Penalty Deep Belief Networks Using Mixed Norm Constraint Summary Deep Belief Networks DBN successfully applied popular machine learning task Specifically applied handwritten digit recognition DBNs achieved approximate accuracy rate 988 effort optimize data representation achieved DBN maximize descriptive p...
[-0.05097030848264694, 0.11358065903186798, -0.018799735233187675, 0.04781665652990341, -0.004256457556039095, 0.031146284192800522, 0.09198739379644394, 0.025534050539135933, 0.028942802920937538, 0.012865393422544003, -0.00016754117677919567, -0.06164489686489105, 0.002961382968351245, 0.015032391063869, -0.006708024...
859
859
['Kyunghyun Cho']
1306.2801v4
In this paper, a simple, general method of adding auxiliary stochastic neurons to a multi-layer perceptron is proposed. It is shown that the proposed method is a generalization of recently successful methods of dropout (Hinton et al., 2012), explicit noise injection (Vincent et al., 2010; Bishop, 1995) and semantic has...
Understanding Dropout: Training Multi-Layer Perceptrons with Auxiliary Independent Stochastic Neurons
2,013
http://arxiv.org/pdf/1306.2801v4
Title Understanding Dropout Training MultiLayer Perceptrons Auxiliary Independent Stochastic Neurons Summary paper simple general method adding auxiliary stochastic neuron multilayer perceptron proposed shown proposed method generalization recently successful method dropout Hinton et al 2012 explicit noise injection Vi...
[-0.0067490264773368835, 0.002243395894765854, -0.0053978790529072285, 0.02976827323436737, -0.007662035990506411, 0.004343375097960234, 0.05553474649786949, -0.04696306958794594, -0.001002069446258247, 0.02266785502433777, -0.010503916069865227, 0.03321674466133118, 0.0028659862000495195, 0.07331717759370804, 0.043902...
860
860
['Linlin Cao', 'Ran He', 'Bao-Gang Hu']
1604.05198v1
This work is a further study on the Generalized Constraint Neural Network (GCNN) model [1], [2]. Two challenges are encountered in the study, that is, to embed any type of prior information and to select its imposing schemes. The work focuses on the second challenge and studies a new constraint imposing scheme for equa...
Locally Imposing Function for Generalized Constraint Neural Networks - A Study on Equality Constraints
2,016
http://arxiv.org/pdf/1604.05198v1
Title Locally Imposing Function Generalized Constraint Neural Networks Study Equality Constraints Summary work study Generalized Constraint Neural Network GCNN model 1 2 Two challenge encountered study embed type prior information select imposing scheme work focus second challenge study new constraint imposing scheme e...
[0.012120881117880344, 0.06214816868305206, -0.016614563763141632, 0.014303338713943958, -0.018026331439614296, -0.07098133116960526, 0.04174258932471275, -0.0015680597862228751, -0.0076708621345460415, 0.01642751693725586, 0.008081131614744663, -0.021789826452732086, -0.00266508967615664, 0.03973434492945671, 0.034251...
861
861
['Gabriel Kronberger', 'Michael Kommenda']
1305.3794v2
In this contribution we describe an approach to evolve composite covariance functions for Gaussian processes using genetic programming. A critical aspect of Gaussian processes and similar kernel-based models such as SVM is, that the covariance function should be adapted to the modeled data. Frequently, the squared expo...
Evolution of Covariance Functions for Gaussian Process Regression using Genetic Programming
2,013
http://arxiv.org/pdf/1305.3794v2
Title Evolution Covariance Functions Gaussian Process Regression using Genetic Programming Summary contribution describe approach evolve composite covariance function Gaussian process using genetic programming critical aspect Gaussian process similar kernelbased model SVM covariance function adapted modeled data Freque...
[-0.03350526839494705, 0.07776971906423569, -0.022528696805238724, -0.0037425400223582983, 0.006588541436940432, 0.022453904151916504, 0.03150999918580055, -0.01943681761622429, -0.011949557811021805, 0.0022121621295809746, 0.04391717538237572, 0.034531835466623306, -0.000879958737641573, 0.1294478476047516, 0.00305702...
862
862
['Nan Wang', 'Jan Melchior', 'Laurenz Wiskott']
1401.5900v1
We present a theoretical analysis of Gaussian-binary restricted Boltzmann machines (GRBMs) from the perspective of density models. The key aspect of this analysis is to show that GRBMs can be formulated as a constrained mixture of Gaussians, which gives a much better insight into the model's capabilities and limitation...
Gaussian-binary Restricted Boltzmann Machines on Modeling Natural Image Statistics
2,014
http://arxiv.org/pdf/1401.5900v1
Title Gaussianbinary Restricted Boltzmann Machines Modeling Natural Image Statistics Summary present theoretical analysis Gaussianbinary restricted Boltzmann machine GRBMs perspective density model key aspect analysis show GRBMs formulated constrained mixture Gaussians give much better insight model capability limitati...
[0.037242915481328964, -0.025194264948368073, -0.02230438031256199, 0.05198498070240021, -0.0034332561772316694, -0.01782248169183731, 0.045624710619449615, 0.02695692516863346, -0.04261785000562668, -0.01116385217756033, -0.01630368083715439, -0.008443581871688366, 0.022364016622304916, 0.0454404279589653, 0.008694514...
863
863
['Siamak Ravanbakhsh', 'Russell Greiner', 'Brendan Frey']
1405.1436v1
A new approach to maximum likelihood learning of discrete graphical models and RBM in particular is introduced. Our method, Perturb and Descend (PD) is inspired by two ideas (I) perturb and MAP method for sampling (II) learning by Contrastive Divergence minimization. In contrast to perturb and MAP, PD leverages trainin...
Training Restricted Boltzmann Machine by Perturbation
2,014
http://arxiv.org/pdf/1405.1436v1
Title Training Restricted Boltzmann Machine Perturbation Summary new approach maximum likelihood learning discrete graphical model RBM particular introduced method Perturb Descend PD inspired two idea perturb MAP method sampling II learning Contrastive Divergence minimization contrast perturb MAP PD leverage training d...
[-0.013479205779731274, -0.030411183834075928, -0.029634486883878708, 0.02206467278301716, -0.012852332554757595, -0.001959493849426508, 0.02294575795531273, 0.00039903982542455196, 0.005045519210398197, 0.001177957165054977, -0.030495693907141685, 0.04076455906033516, -0.002123118145391345, 0.020608248189091682, 0.038...
864
864
['Xiao-Lei Zhang']
1408.0848v8
Multilayer bootstrap network builds a gradually narrowed multilayer nonlinear network from bottom up for unsupervised nonlinear dimensionality reduction. Each layer of the network is a nonparametric density estimator. It consists of a group of k-centroids clusterings. Each clustering randomly selects data points with r...
Multilayer bootstrap networks
2,014
http://arxiv.org/pdf/1408.0848v8
Title Multilayer bootstrap network Summary Multilayer bootstrap network build gradually narrowed multilayer nonlinear network bottom unsupervised nonlinear dimensionality reduction layer network nonparametric density estimator consists group kcentroids clustering clustering randomly selects data point randomly selected...
[-0.04715077579021454, 0.016762183979153633, -0.039488036185503006, -0.005515534430742264, -0.037119973450899124, -0.019222823902964592, 0.04221014678478241, -0.003004320664331317, -0.06774306297302246, 0.051473766565322876, -0.024001939222216606, 0.024524623528122902, 0.015463277697563171, 0.025538334622979164, 0.0409...
865
865
['Sergey Demyanov', 'James Bailey', 'Ramamohanarao Kotagiri', 'Christopher Leckie']
1502.04434v3
In many classification problems a classifier should be robust to small variations in the input vector. This is a desired property not only for particular transformations, such as translation and rotation in image classification problems, but also for all others for which the change is small enough to retain the object ...
Invariant backpropagation: how to train a transformation-invariant neural network
2,015
http://arxiv.org/pdf/1502.04434v3
Title Invariant backpropagation train transformationinvariant neural network Summary many classification problem classifier robust small variation input vector desired property particular transformation translation rotation image classification problem also others change small enough retain object perceptually indistin...
[0.015735244378447533, 0.03271784633398056, -0.03202942758798599, 0.018016867339611053, 0.02261238917708397, -0.018340522423386574, 0.025017978623509407, 0.0011434972984716296, -0.05588329955935478, -0.028222892433404922, 0.0007722434820607305, 0.03622652217745781, 0.030474750325083733, 0.004983484745025635, 0.05090219...
866
866
['Danila Doroshin', 'Alexander Yamshinin', 'Nikolay Lubimov', 'Marina Nastasenko', 'Mikhail Kotov', 'Maxim Tkachenko']
1503.05471v1
This paper presents a novel approach to speaker subspace modelling based on Gaussian-Binary Restricted Boltzmann Machines (GRBM). The proposed model is based on the idea of shared factors as in the Probabilistic Linear Discriminant Analysis (PLDA). GRBM hidden layer is divided into speaker and channel factors, herein t...
Shared latent subspace modelling within Gaussian-Binary Restricted Boltzmann Machines for NIST i-Vector Challenge 2014
2,015
http://arxiv.org/pdf/1503.05471v1
Title Shared latent subspace modelling within GaussianBinary Restricted Boltzmann Machines NIST iVector Challenge 2014 Summary paper present novel approach speaker subspace modelling based GaussianBinary Restricted Boltzmann Machines GRBM proposed model based idea shared factor Probabilistic Linear Discriminant Analysi...
[0.019772611558437347, 0.007206613197922707, -0.02559203840792179, 0.08306772261857986, -0.011368933133780956, 0.014688745141029358, 0.029226558282971382, -0.027302538976073265, -0.04247799888253212, 0.02238532342016697, -0.05575773864984512, -0.03574942797422409, 0.03611341491341591, 0.017466768622398376, 0.0057443357...
867
867
['Sebastian Urban', 'Patrick van der Smagt']
1503.05724v3
Existing approaches to combine both additive and multiplicative neural units either use a fixed assignment of operations or require discrete optimization to determine what function a neuron should perform. This leads either to an inefficient distribution of computational resources or an extensive increase in the comput...
A Neural Transfer Function for a Smooth and Differentiable Transition Between Additive and Multiplicative Interactions
2,015
http://arxiv.org/pdf/1503.05724v3
Title Neural Transfer Function Smooth Differentiable Transition Additive Multiplicative Interactions Summary Existing approach combine additive multiplicative neural unit either use fixed assignment operation require discrete optimization determine function neuron perform lead either inefficient distribution computatio...
[-0.02768981270492077, 0.05566762387752533, 0.0052767000161111355, 0.03848479315638542, 0.04850592091679573, -0.008199125528335571, 0.028266819193959236, -0.0469738207757473, -0.0611237995326519, 0.005652204621583223, -0.024118052795529366, -0.042379822582006454, 0.04574961215257645, 0.06665749847888947, 0.034806590527...
868
868
['Ankit B. Patel', 'Tan Nguyen', 'Richard G. Baraniuk']
1612.01936v1
We develop a probabilistic framework for deep learning based on the Deep Rendering Mixture Model (DRMM), a new generative probabilistic model that explicitly capture variations in data due to latent task nuisance variables. We demonstrate that max-sum inference in the DRMM yields an algorithm that exactly reproduces th...
A Probabilistic Framework for Deep Learning
2,016
http://arxiv.org/pdf/1612.01936v1
Title Probabilistic Framework Deep Learning Summary develop probabilistic framework deep learning based Deep Rendering Mixture Model DRMM new generative probabilistic model explicitly capture variation data due latent task nuisance variable demonstrate maxsum inference DRMM yield algorithm exactly reproduces operation ...
[-0.023877430707216263, 0.028987117111682892, -0.0010197419906035066, 0.02622407302260399, 0.01665334403514862, -0.02270248532295227, 0.027868950739502907, -0.021976148709654808, -0.0785486102104187, 0.040055111050605774, 0.025343673303723335, 0.004170405212789774, 0.017193099483847618, 0.06400845944881439, 0.031361274...
869
869
['Timothy J. Draelos', 'Nadine E. Miner', 'Christopher C. Lamb', 'Jonathan A. Cox', 'Craig M. Vineyard', 'Kristofor D. Carlson', 'William M. Severa', 'Conrad D. James', 'James B. Aimone']
1612.03770v2
Neural machine learning methods, such as deep neural networks (DNN), have achieved remarkable success in a number of complex data processing tasks. These methods have arguably had their strongest impact on tasks such as image and audio processing - data processing domains in which humans have long held clear advantages...
Neurogenesis Deep Learning
2,016
http://arxiv.org/pdf/1612.03770v2
Title Neurogenesis Deep Learning Summary Neural machine learning method deep neural network DNN achieved remarkable success number complex data processing task method arguably strongest impact task image audio processing data processing domain human long held clear advantage conventional algorithm contrast biological n...
[0.007181879132986069, 0.07644831389188766, -0.030216600745916367, -0.029585018754005432, 0.05933033302426338, 0.008174392394721508, 0.058180071413517, 0.01502329669892788, -0.03695451468229294, 0.041550710797309875, 0.0292380228638649, 0.0026918291114270687, 0.00036305346293374896, 0.044411759823560715, 0.032330386340...
870
870
['Sergey M. Plis', 'Devon R. Hjelm', 'Ruslan Salakhutdinov', 'Vince D. Calhoun']
1312.5847v3
Deep learning methods have recently made notable advances in the tasks of classification and representation learning. These tasks are important for brain imaging and neuroscience discovery, making the methods attractive for porting to a neuroimager's toolbox. Success of these methods is, in part, explained by the flexi...
Deep learning for neuroimaging: a validation study
2,013
http://arxiv.org/pdf/1312.5847v3
Title Deep learning neuroimaging validation study Summary Deep learning method recently made notable advance task classification representation learning task important brain imaging neuroscience discovery making method attractive porting neuroimagers toolbox Success method part explained flexibility deep learning model...
[-0.01837492361664772, 0.01100934948772192, -0.02576216496527195, -0.012718332931399345, 0.02512184903025627, 0.015584123320877552, 0.04696991294622421, 0.023571521043777466, -6.270770245464519e-05, 0.053926657885313034, -0.05788812041282654, -0.022314229980111122, 0.04795408248901367, 0.06083568558096886, 0.0602912306...
871
871
['Jost Tobias Springenberg', 'Martin Riedmiller']
1312.6116v2
We present a probabilistic variant of the recently introduced maxout unit. The success of deep neural networks utilizing maxout can partly be attributed to favorable performance under dropout, when compared to rectified linear units. It however also depends on the fact that each maxout unit performs a pooling operation...
Improving Deep Neural Networks with Probabilistic Maxout Units
2,013
http://arxiv.org/pdf/1312.6116v2
Title Improving Deep Neural Networks Probabilistic Maxout Units Summary present probabilistic variant recently introduced maxout unit success deep neural network utilizing maxout partly attributed favorable performance dropout compared rectified linear unit however also depends fact maxout unit performs pooling operati...
[-0.005381505936384201, 0.019057119265198708, 0.016910037025809288, 0.03946641832590103, -0.018241185694932938, 0.012659365311264992, 0.048342134803533554, -0.019333038479089737, -0.04003531485795975, 0.022451849654316902, 0.011953772976994514, -0.03398923575878143, 0.002058961894363165, 0.07588708400726318, 0.02126079...
872
872
['Fabrice Rossi']
1407.0611v1
In numerous applicative contexts, data are too rich and too complex to be represented by numerical vectors. A general approach to extend machine learning and data mining techniques to such data is to really on a dissimilarity or on a kernel that measures how different or similar two objects are. This approach has been ...
How Many Dissimilarity/Kernel Self Organizing Map Variants Do We Need?
2,014
http://arxiv.org/pdf/1407.0611v1
Title Many DissimilarityKernel Self Organizing Map Variants Need Summary numerous applicative context data rich complex represented numerical vector general approach extend machine learning data mining technique data really dissimilarity kernel measure different similar two object approach used define several variant S...
[-0.03707399219274521, -0.017206881195306778, -0.04620763659477234, 0.03050832822918892, -0.014078105799853802, -0.007796098478138447, 0.07212647050619125, -0.014766501262784004, 0.0195127222687006, -0.016467800363898277, -0.03621876239776611, 0.07111173868179321, -0.01064393576234579, 0.002003050409257412, -0.02756491...
873
873
['John R. Hershey', 'Jonathan Le Roux', 'Felix Weninger']
1409.2574v4
Model-based methods and deep neural networks have both been tremendously successful paradigms in machine learning. In model-based methods, problem domain knowledge can be built into the constraints of the model, typically at the expense of difficulties during inference. In contrast, deterministic deep neural networks a...
Deep Unfolding: Model-Based Inspiration of Novel Deep Architectures
2,014
http://arxiv.org/pdf/1409.2574v4
Title Deep Unfolding ModelBased Inspiration Novel Deep Architectures Summary Modelbased method deep neural network tremendously successful paradigm machine learning modelbased method problem domain knowledge built constraint model typically expense difficulty inference contrast deterministic deep neural network constru...
[-0.04810643196105957, 0.036373239010572433, 0.010821335017681122, 0.016885582357645035, 0.01888921670615673, -0.050883520394563675, 0.04695967584848404, -0.012539033778011799, -0.07194012403488159, 0.028032740578055382, -0.024332191795110703, 0.010094077326357365, 0.02667004056274891, 0.07313326746225357, 0.0220590103...
874
874
['Niklas Wahlström', 'Thomas B. Schön', 'Marc Peter Deisenroth']
1410.7550v1
Modeling dynamical systems is important in many disciplines, e.g., control, robotics, or neurotechnology. Commonly the state of these systems is not directly observed, but only available through noisy and potentially high-dimensional observations. In these cases, system identification, i.e., finding the measurement map...
Learning deep dynamical models from image pixels
2,014
http://arxiv.org/pdf/1410.7550v1
Title Learning deep dynamical model image pixel Summary Modeling dynamical system important many discipline eg control robotics neurotechnology Commonly state system directly observed available noisy potentially highdimensional observation case system identification ie finding measurement mapping transition mapping sys...
[-0.044298529624938965, 0.030958494171500206, -0.03170099854469299, 0.05849612504243851, -0.0041681816801428795, -0.015608853660523891, 0.04241286963224411, -0.0038738218136131763, -0.06255733221769333, 0.014287787489593029, 0.044057298451662064, 0.01682785153388977, -0.003498351201415062, 0.08716800808906555, 0.041112...
875
875
['Harri Valpola']
1411.7783v2
A network supporting deep unsupervised learning is presented. The network is an autoencoder with lateral shortcut connections from the encoder to decoder at each level of the hierarchy. The lateral shortcut connections allow the higher levels of the hierarchy to focus on abstract invariant features. While standard auto...
From neural PCA to deep unsupervised learning
2,014
http://arxiv.org/pdf/1411.7783v2
Title neural PCA deep unsupervised learning Summary network supporting deep unsupervised learning presented network autoencoder lateral shortcut connection encoder decoder level hierarchy lateral shortcut connection allow higher level hierarchy focus abstract invariant feature standard autoencoders analogous latent var...
[-0.020638780668377876, 0.08852510899305344, -0.011860786005854607, 0.04436541721224785, 0.026439804583787918, 0.018264351412653923, 0.06064463034272194, -0.012935417704284191, -0.04578427970409393, 0.012000597082078457, 0.023024294525384903, 0.0461929589509964, -0.006049367133527994, 0.11999958753585815, 0.02286304533...
876
876
['Ian J. Goodfellow', 'Oriol Vinyals', 'Andrew M. Saxe']
1412.6544v6
Training neural networks involves solving large-scale non-convex optimization problems. This task has long been believed to be extremely difficult, with fear of local minima and other obstacles motivating a variety of schemes to improve optimization, such as unsupervised pretraining. However, modern neural networks are...
Qualitatively characterizing neural network optimization problems
2,014
http://arxiv.org/pdf/1412.6544v6
Title Qualitatively characterizing neural network optimization problem Summary Training neural network involves solving largescale nonconvex optimization problem task long believed extremely difficult fear local minimum obstacle motivating variety scheme improve optimization unsupervised pretraining However modern neur...
[-0.026326825842261314, 0.024307403713464737, -0.014821956865489483, 0.03130592033267021, -0.016995619982481003, -0.03342808410525322, 0.008735844865441322, -0.001642549759708345, -0.034666623920202255, 0.01714233122766018, 0.011867056600749493, 0.031946271657943726, -0.0019691416528075933, 0.02467413991689682, 0.04791...
877
877
['Arnab Paul', 'Suresh Venkatasubramanian']
1412.6621v3
Why does Deep Learning work? What representations does it capture? How do higher-order representations emerge? We study these questions from the perspective of group theory, thereby opening a new approach towards a theory of Deep learning. One factor behind the recent resurgence of the subject is a key algorithmic st...
Why does Deep Learning work? - A perspective from Group Theory
2,014
http://arxiv.org/pdf/1412.6621v3
Title Deep Learning work perspective Group Theory Summary Deep Learning work representation capture higherorder representation emerge study question perspective group theory thereby opening new approach towards theory Deep learning One factor behind recent resurgence subject key algorithmic step called pretraining firs...
[-0.017468933016061783, 0.08319175988435745, -0.036700539290905, -0.01308742817491293, -0.01370431762188673, 0.003490353235974908, 0.08844083547592163, 0.004378741607069969, -0.08824407309293747, 0.004143625032156706, 0.019065510481595993, -0.03476626053452492, -0.005053828004747629, 0.03994399309158325, 0.083658069372...
878
878
['Caglar Gulcehre', 'Marcin Moczulski', 'Yoshua Bengio']
1412.7419v5
Stochastic gradient algorithms have been the main focus of large-scale learning problems and they led to important successes in machine learning. The convergence of SGD depends on the careful choice of learning rate and the amount of the noise in stochastic estimates of the gradients. In this paper, we propose a new ad...
ADASECANT: Robust Adaptive Secant Method for Stochastic Gradient
2,014
http://arxiv.org/pdf/1412.7419v5
Title ADASECANT Robust Adaptive Secant Method Stochastic Gradient Summary Stochastic gradient algorithm main focus largescale learning problem led important success machine learning convergence SGD depends careful choice learning rate amount noise stochastic estimate gradient paper propose new adaptive learning rate al...
[0.009424963966012001, -0.013777581974864006, -0.00192730943672359, 0.027817241847515106, 0.014022969640791416, -0.02599353902041912, 0.032566990703344345, -0.04041124880313873, -0.0035957740619778633, 0.03253929316997528, -0.001778207952156663, -0.020874634385108948, 0.022946560755372047, 0.029129309579730034, 0.00011...
879
879
['Yongxin Yang', 'Timothy M. Hospedales']
1412.7489v3
In this paper, we provide a new neural-network based perspective on multi-task learning (MTL) and multi-domain learning (MDL). By introducing the concept of a semantic descriptor, this framework unifies MDL and MTL as well as encompassing various classic and recent MTL/MDL algorithms by interpreting them as different w...
A Unified Perspective on Multi-Domain and Multi-Task Learning
2,014
http://arxiv.org/pdf/1412.7489v3
Title Unified Perspective MultiDomain MultiTask Learning Summary paper provide new neuralnetwork based perspective multitask learning MTL multidomain learning MDL introducing concept semantic descriptor framework unifies MDL MTL well encompassing various classic recent MTLMDL algorithm interpreting different way constr...
[-0.03835504502058029, -0.0007626500446349382, 0.006105609238147736, 0.002690607914701104, 0.00775135587900877, 0.02102486789226532, 0.08897290378808975, -0.029509609565138817, -0.028897147625684738, -0.0681108683347702, -0.0457310825586319, -0.033412013202905655, -0.03511973097920418, 0.029965292662382126, 0.031980104...
880
880
['Robert A. Murphy']
1501.07227v5
The random cluster model is used to define an upper bound on a distance measure as a function of the number of data points to be classified and the expected value of the number of classes to form in a hybrid K-means and regression classification methodology, with the intent of detecting anomalies. Conditions are given ...
A Neural Network Anomaly Detector Using the Random Cluster Model
2,015
http://arxiv.org/pdf/1501.07227v5
Title Neural Network Anomaly Detector Using Random Cluster Model Summary random cluster model used define upper bound distance measure function number data point classified expected value number class form hybrid Kmeans regression classification methodology intent detecting anomaly Conditions given identification class...
[0.002476252382621169, -0.014347896911203861, -0.020285645499825478, 0.05037723854184151, -0.021373439580202103, -0.014406777918338776, 0.02718907780945301, 0.016552196815609932, 0.02085110917687416, 0.014222637750208378, 0.10309093445539474, 0.022898592054843903, -0.01529272273182869, 0.05098186060786247, 0.0070929424...
881
881
['Arnab Paul', 'Suresh Venkatasubramanian']
1504.02462v3
Why does Deep Learning work? What representations does it capture? How do higher-order representations emerge? We study these questions from the perspective of group theory, thereby opening a new approach towards a theory of Deep learning. One factor behind the recent resurgence of the subject is a key algorithmic st...
A Group Theoretic Perspective on Unsupervised Deep Learning
2,015
http://arxiv.org/pdf/1504.02462v3
Title Group Theoretic Perspective Unsupervised Deep Learning Summary Deep Learning work representation capture higherorder representation emerge study question perspective group theory thereby opening new approach towards theory Deep learning One factor behind recent resurgence subject key algorithmic step called em pr...
[-0.009884160943329334, 0.07413185387849808, -0.028366418555378914, -0.0016227675369009376, -0.008837842382490635, 0.014756530523300171, 0.07110969722270966, -0.017646580934524536, -0.0825798287987709, 0.0019772634841501713, -0.0016793075483292341, -0.04635702818632126, -0.008833921514451504, 0.03528766334056854, 0.081...
882
882
['Yunchen Pu', 'Xin Yuan', 'Lawrence Carin']
1504.04054v1
A generative model is developed for deep (multi-layered) convolutional dictionary learning. A novel probabilistic pooling operation is integrated into the deep model, yielding efficient bottom-up (pretraining) and top-down (refinement) probabilistic learning. Experimental results demonstrate powerful capabilities of th...
A Generative Model for Deep Convolutional Learning
2,015
http://arxiv.org/pdf/1504.04054v1
Title Generative Model Deep Convolutional Learning Summary generative model developed deep multilayered convolutional dictionary learning novel probabilistic pooling operation integrated deep model yielding efficient bottomup pretraining topdown refinement probabilistic learning Experimental result demonstrate powerful...
[-0.015482412651181221, 0.014986480586230755, -0.0065272171050310135, 0.0221563708037138, -0.005253560375422239, -0.02257644571363926, 0.043189022690057755, -0.03754134103655815, -0.02604890801012516, 0.002766217105090618, 0.05996289104223251, 0.02258172072470188, -0.0118293147534132, 0.07253018766641617, 0.03399497643...
883
883
['Zhiyuan Tang', 'Dong Wang', 'Yiqiao Pan', 'Zhiyong Zhang']
1506.02256v1
Pre-training is crucial for learning deep neural networks. Most of existing pre-training methods train simple models (e.g., restricted Boltzmann machines) and then stack them layer by layer to form the deep structure. This layer-wise pre-training has found strong theoretical foundation and broad empirical support. Howe...
Knowledge Transfer Pre-training
2,015
http://arxiv.org/pdf/1506.02256v1
Title Knowledge Transfer Pretraining Summary Pretraining crucial learning deep neural network existing pretraining method train simple model eg restricted Boltzmann machine stack layer layer form deep structure layerwise pretraining found strong theoretical foundation broad empirical support However easy employ method ...
[-0.012281407602131367, 0.05470841005444527, 0.009022770449519157, 0.024301918223500252, 0.02616831474006176, -0.01433711126446724, 0.016306878998875618, -0.03301497548818588, -0.04225548729300499, -0.01741214282810688, -0.0617535226047039, 0.03170616924762726, -0.0021883330773562193, 0.028970902785658836, 0.0057556550...
884
884
['Junbo Zhao', 'Michael Mathieu', 'Ross Goroshin', 'Yann LeCun']
1506.02351v8
We present a novel architecture, the "stacked what-where auto-encoders" (SWWAE), which integrates discriminative and generative pathways and provides a unified approach to supervised, semi-supervised and unsupervised learning without relying on sampling during training. An instantiation of SWWAE uses a convolutional ne...
Stacked What-Where Auto-encoders
2,015
http://arxiv.org/pdf/1506.02351v8
Title Stacked WhatWhere Autoencoders Summary present novel architecture stacked whatwhere autoencoders SWWAE integrates discriminative generative pathway provides unified approach supervised semisupervised unsupervised learning without relying sampling training instantiation SWWAE us convolutional net Convnet LeCun et ...
[-0.01981442980468273, 0.08821431547403336, -0.01798466593027115, 0.053672514855861664, -0.0015582708874717355, 0.007265186402946711, 0.06560763716697693, -0.007751486264169216, -0.011257002130150795, -0.020326735451817513, -0.043799884617328644, 0.010031813755631447, -0.005495553836226463, 0.05339856073260307, 0.05001...
885
885
['Yann Ollivier', 'Corentin Tallec', 'Guillaume Charpiat']
1507.07680v2
We introduce the "NoBackTrack" algorithm to train the parameters of dynamical systems such as recurrent neural networks. This algorithm works in an online, memoryless setting, thus requiring no backpropagation through time, and is scalable, avoiding the large computational and memory cost of maintaining the full gradie...
Training recurrent networks online without backtracking
2,015
http://arxiv.org/pdf/1507.07680v2
Title Training recurrent network online without backtracking Summary introduce NoBackTrack algorithm train parameter dynamical system recurrent neural network algorithm work online memoryless setting thus requiring backpropagation time scalable avoiding large computational memory cost maintaining full gradient current ...
[-0.010960999876260757, 0.04125826060771942, 0.006236083339899778, -0.0005737123428843915, 0.001664411392994225, -0.019761379808187485, -0.024306312203407288, -0.019674770534038544, -0.020873045548796654, -0.006680076476186514, 0.009317561984062195, 0.023852698504924774, 0.02771538496017456, 0.055159419775009155, 0.042...
886
886
['John R. Hershey', 'Zhuo Chen', 'Jonathan Le Roux', 'Shinji Watanabe']
1508.04306v1
We address the problem of acoustic source separation in a deep learning framework we call "deep clustering." Rather than directly estimating signals or masking functions, we train a deep network to produce spectrogram embeddings that are discriminative for partition labels given in training data. Previous deep network ...
Deep clustering: Discriminative embeddings for segmentation and separation
2,015
http://arxiv.org/pdf/1508.04306v1
Title Deep clustering Discriminative embeddings segmentation separation Summary address problem acoustic source separation deep learning framework call deep clustering Rather directly estimating signal masking function train deep network produce spectrogram embeddings discriminative partition label given training data ...
[-0.031613484025001526, 0.0043369196355342865, 0.003046144498512149, 0.059624943882226944, -0.002430841326713562, -0.03009999915957451, 0.0710962787270546, 0.001544583821669221, -0.053623396903276443, 0.010651652701199055, -0.062092483043670654, 0.024329975247383118, 0.0025264392606914043, 0.01707950234413147, 0.019115...
887
887
['Aren Jansen', 'Gregory Sell', 'Vince Lyzinski']
1508.04422v3
Several popular graph embedding techniques for representation learning and dimensionality reduction rely on performing computationally expensive eigendecompositions to derive a nonlinear transformation of the input data space. The resulting eigenvectors encode the embedding coordinates for the training samples only, an...
Scalable Out-of-Sample Extension of Graph Embeddings Using Deep Neural Networks
2,015
http://arxiv.org/pdf/1508.04422v3
Title Scalable OutofSample Extension Graph Embeddings Using Deep Neural Networks Summary Several popular graph embedding technique representation learning dimensionality reduction rely performing computationally expensive eigendecompositions derive nonlinear transformation input data space resulting eigenvectors encode...
[-0.026009386405348778, -0.004638715647161007, -0.00652187317609787, 0.007820820435881615, 0.02224012091755867, 0.008739415556192398, 0.0071080937050282955, -0.002112424233928323, 0.03175557032227516, 0.03627576306462288, 0.00741588743403554, 0.05737767741084099, -0.005602273158729076, 0.007984697818756104, 0.049217622...
888
888
['Suyog Gupta', 'Wei Zhang', 'Fei Wang']
1509.04210v3
This paper presents Rudra, a parameter server based distributed computing framework tuned for training large-scale deep neural networks. Using variants of the asynchronous stochastic gradient descent algorithm we study the impact of synchronization protocol, stale gradient updates, minibatch size, learning rates, and n...
Model Accuracy and Runtime Tradeoff in Distributed Deep Learning:A Systematic Study
2,015
http://arxiv.org/pdf/1509.04210v3
Title Model Accuracy Runtime Tradeoff Distributed Deep LearningA Systematic Study Summary paper present Rudra parameter server based distributed computing framework tuned training largescale deep neural network Using variant asynchronous stochastic gradient descent algorithm study impact synchronization protocol stale ...
[0.0023449964355677366, 0.04158483445644379, 0.010681022889912128, 0.05070742592215538, 0.0064909085631370544, 0.0016931610880419612, 0.06247880682349205, -0.03652351349592209, -0.039995741099119186, 0.03298861160874367, -0.024701006710529327, -0.010304884985089302, 0.02436564676463604, 0.03155328333377838, -0.04015511...
889
889
['David Duvenaud', 'Dougal Maclaurin', 'Jorge Aguilera-Iparraguirre', 'Rafael Gómez-Bombarelli', 'Timothy Hirzel', 'Alán Aspuru-Guzik', 'Ryan P. Adams']
1509.09292v2
We introduce a convolutional neural network that operates directly on graphs. These networks allow end-to-end learning of prediction pipelines whose inputs are graphs of arbitrary size and shape. The architecture we present generalizes standard molecular feature extraction methods based on circular fingerprints. We sho...
Convolutional Networks on Graphs for Learning Molecular Fingerprints
2,015
http://arxiv.org/pdf/1509.09292v2
Title Convolutional Networks Graphs Learning Molecular Fingerprints Summary introduce convolutional neural network operates directly graph network allow endtoend learning prediction pipeline whose input graph arbitrary size shape architecture present generalizes standard molecular feature extraction method based circul...
[0.02733602561056614, 0.013026212342083454, -0.02836785279214382, 0.0018979154992848635, -0.014865434728562832, -0.01779913902282715, 0.04825456440448761, 0.017269223928451538, 0.02236657217144966, 0.035699907690286636, 0.05706275627017021, 0.03128650039434433, -0.021096352487802505, 0.10005316138267517, 0.033814862370...
890
890
['Oswin Krause', 'Asja Fischer', 'Christian Igel']
1510.01624v4
Estimating the log-likelihood gradient with respect to the parameters of a Restricted Boltzmann Machine (RBM) typically requires sampling using Markov Chain Monte Carlo (MCMC) techniques. To save computation time, the Markov chains are only run for a small number of steps, which leads to a biased estimate. This bias ca...
Population-Contrastive-Divergence: Does Consistency help with RBM training?
2,015
http://arxiv.org/pdf/1510.01624v4
Title PopulationContrastiveDivergence Consistency help RBM training Summary Estimating loglikelihood gradient respect parameter Restricted Boltzmann Machine RBM typically requires sampling using Markov Chain Monte Carlo MCMC technique save computation time Markov chain run small number step lead biased estimate bias ca...
[0.01718243584036827, -0.009649938903748989, -0.010894007980823517, -0.009393826127052307, -0.006687752902507782, -0.025903603062033653, 0.034462735056877136, -0.005490076262503862, -0.05730156973004341, 0.031931161880493164, 0.008940186351537704, 0.0010358005529269576, 0.04077569395303726, 0.005169510841369629, 0.0228...
891
891
['Izhar Wallach', 'Michael Dzamba', 'Abraham Heifets']
1510.02855v1
Deep convolutional neural networks comprise a subclass of deep neural networks (DNN) with a constrained architecture that leverages the spatial and temporal structure of the domain they model. Convolutional networks achieve the best predictive performance in areas such as speech and image recognition by hierarchically ...
AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery
2,015
http://arxiv.org/pdf/1510.02855v1
Title AtomNet Deep Convolutional Neural Network Bioactivity Prediction Structurebased Drug Discovery Summary Deep convolutional neural network comprise subclass deep neural network DNN constrained architecture leverage spatial temporal structure domain model Convolutional network achieve best predictive performance are...
[0.010271945968270302, 0.053097814321517944, -0.019327670335769653, 0.025630924850702286, 0.045571018010377884, -0.03217602148652077, 0.06780460476875305, 0.019847920164465904, 0.048916690051555634, 0.017969787120819092, -0.04063716530799866, 0.03772808238863945, -0.016215762123465538, 0.06311431527137756, 0.0283563230...
892
892
['Nicolas Papernot', 'Patrick McDaniel', 'Xi Wu', 'Somesh Jha', 'Ananthram Swami']
1511.04508v2
Deep learning algorithms have been shown to perform extremely well on many classical machine learning problems. However, recent studies have shown that deep learning, like other machine learning techniques, is vulnerable to adversarial samples: inputs crafted to force a deep neural network (DNN) to provide adversary-se...
Distillation as a Defense to Adversarial Perturbations against Deep Neural Networks
2,015
http://arxiv.org/pdf/1511.04508v2
Title Distillation Defense Adversarial Perturbations Deep Neural Networks Summary Deep learning algorithm shown perform extremely well many classical machine learning problem However recent study shown deep learning like machine learning technique vulnerable adversarial sample input crafted force deep neural network DN...
[0.00978070218116045, 0.056645773351192474, -0.016303276643157005, 0.007887858897447586, -0.03942437097430229, -0.020568089559674263, 0.04720168560743332, -0.01884683221578598, -0.028821565210819244, -0.017682690173387527, 0.023345734924077988, 0.04580553621053696, -0.0046011218801140785, 0.03757333382964134, 0.0346834...
893
893
['Dustin Tran', 'Rajesh Ranganath', 'David M. Blei']
1511.06499v4
Variational inference is a powerful tool for approximate inference, and it has been recently applied for representation learning with deep generative models. We develop the variational Gaussian process (VGP), a Bayesian nonparametric variational family, which adapts its shape to match complex posterior distributions. T...
The Variational Gaussian Process
2,015
http://arxiv.org/pdf/1511.06499v4
Title Variational Gaussian Process Summary Variational inference powerful tool approximate inference recently applied representation learning deep generative model develop variational Gaussian process VGP Bayesian nonparametric variational family adapts shape match complex posterior distribution VGP generates approxima...
[-0.020552849397063255, 0.09730485081672668, -0.01959320902824402, -0.021689001470804214, -0.012611591257154942, -0.008213704451918602, 0.015472936443984509, -0.02307627536356449, -0.08204922080039978, 0.02722841314971447, 0.01838023029267788, -0.0056540523655712605, 0.0158979631960392, 0.08420240879058838, 0.052530273...
894
894
['Ilia Zintchenko', 'Matthew Hastings', 'Nathan Wiebe', 'Ethan Brown', 'Matthias Troyer']
1512.03025v1
Heuristic optimisers which search for an optimal configuration of variables relative to an objective function often get stuck in local optima where the algorithm is unable to find further improvement. The standard approach to circumvent this problem involves periodically restarting the algorithm from random initial con...
Partial Reinitialisation for Optimisers
2,015
http://arxiv.org/pdf/1512.03025v1
Title Partial Reinitialisation Optimisers Summary Heuristic optimisers search optimal configuration variable relative objective function often get stuck local optimum algorithm unable find improvement standard approach circumvent problem involves periodically restarting algorithm random initial configuration improvemen...
[-0.037638649344444275, -0.0100403456017375, -0.012705590575933456, -0.029467321932315826, 0.01377557497471571, -0.05436856299638748, -0.03977998346090317, 0.008652768097817898, -0.03583249822258949, -0.031532831490039825, 0.06650076061487198, 0.03275195509195328, -0.010251748375594616, 0.05194227397441864, -0.02478437...
895
895
['Ronen Basri', 'David Jacobs']
1602.04723v1
We consider the ability of deep neural networks to represent data that lies near a low-dimensional manifold in a high-dimensional space. We show that deep networks can efficiently extract the intrinsic, low-dimensional coordinates of such data. We first show that the first two layers of a deep network can exactly embed...
Efficient Representation of Low-Dimensional Manifolds using Deep Networks
2,016
http://arxiv.org/pdf/1602.04723v1
Title Efficient Representation LowDimensional Manifolds using Deep Networks Summary consider ability deep neural network represent data lie near lowdimensional manifold highdimensional space show deep network efficiently extract intrinsic lowdimensional coordinate data first show first two layer deep network exactly em...
[-0.08361966162919998, 0.02978036180138588, -0.018027890473604202, 0.045523494482040405, -0.05180705338716507, 0.012776571325957775, 0.04660985991358757, 0.000518214306794107, 0.020451143383979797, 0.00042285610106773674, 0.00424810079857707, 0.040981099009513855, -0.025472229346632957, 0.059133823961019516, 0.06639036...
896
896
['André L. V. Coelho', 'Fabrício O. de França']
1603.06859v1
Perceptrons are neuronal devices capable of fully discriminating linearly separable classes. Although straightforward to implement and train, their applicability is usually hindered by non-trivial requirements imposed by real-world classification problems. Therefore, several approaches, such as kernel perceptrons, have...
Enhanced perceptrons using contrastive biclusters
2,016
http://arxiv.org/pdf/1603.06859v1
Title Enhanced perceptrons using contrastive biclusters Summary Perceptrons neuronal device capable fully discriminating linearly separable class Although straightforward implement train applicability usually hindered nontrivial requirement imposed realworld classification problem Therefore several approach kernel perc...
[-0.025995424017310143, -0.018527960404753685, -0.04075160250067711, 0.07005468755960464, -0.03558843210339546, 0.002725470345467329, 0.07659794390201569, 0.009443908929824829, 0.06532938033342361, -0.012848589569330215, -0.04604990780353546, 0.011575860902667046, -0.01611573062837124, 0.020199354737997055, 0.018403012...
897
897
['Tetsuya Sakurai', 'Akira Imakura', 'Yuto Inoue', 'Yasunori Futamura']
1605.04639v1
The backpropagation algorithm for calculating gradients has been widely used in computation of weights for deep neural networks (DNNs). This method requires derivatives of objective functions and has some difficulties finding appropriate parameters such as learning rate. In this paper, we propose a novel approach for c...
Alternating optimization method based on nonnegative matrix factorizations for deep neural networks
2,016
http://arxiv.org/pdf/1605.04639v1
Title Alternating optimization method based nonnegative matrix factorization deep neural network Summary backpropagation algorithm calculating gradient widely used computation weight deep neural network DNNs method requires derivative objective function difficulty finding appropriate parameter learning rate paper propo...
[-0.02732893079519272, 0.02925928123295307, -0.006862843409180641, 0.055406827479600906, 0.027116430923342705, -0.028970245271921158, 0.01844571717083454, -0.021081620827317238, -0.022841472178697586, -0.01237912755459547, -0.07110622525215149, -0.017962569370865822, -0.008172176778316498, 0.041417695581912994, 0.03918...
898
898
['Jure Sokolic', 'Raja Giryes', 'Guillermo Sapiro', 'Miguel R. D. Rodrigues']
1605.08254v3
The generalization error of deep neural networks via their classification margin is studied in this work. Our approach is based on the Jacobian matrix of a deep neural network and can be applied to networks with arbitrary non-linearities and pooling layers, and to networks with different architectures such as feed forw...
Robust Large Margin Deep Neural Networks
2,016
http://arxiv.org/pdf/1605.08254v3
Title Robust Large Margin Deep Neural Networks Summary generalization error deep neural network via classification margin studied work approach based Jacobian matrix deep neural network applied network arbitrary nonlinearities pooling layer network different architecture feed forward network residual network analysis l...
[-0.011793049052357674, 0.015552207827568054, -0.02487330697476864, 0.05208481103181839, 0.04645862057805061, -0.029161689803004265, 0.027501219883561134, -0.020609071478247643, -0.0381009578704834, 0.03582538664340973, -0.019938170909881592, 0.03781475871801376, -0.0024499783758074045, 0.018411971628665924, 0.01985204...
899
899
['Daniel Soudry', 'Yair Carmon']
1605.08361v2
We use smoothed analysis techniques to provide guarantees on the training loss of Multilayer Neural Networks (MNNs) at differentiable local minima. Specifically, we examine MNNs with piecewise linear activation functions, quadratic loss and a single output, under mild over-parametrization. We prove that for a MNN with ...
No bad local minima: Data independent training error guarantees for multilayer neural networks
2,016
http://arxiv.org/pdf/1605.08361v2
Title bad local minimum Data independent training error guarantee multilayer neural network Summary use smoothed analysis technique provide guarantee training loss Multilayer Neural Networks MNNs differentiable local minimum Specifically examine MNNs piecewise linear activation function quadratic loss single output mil...
[-0.02413204498589039, 0.04211192950606346, -0.0025384053587913513, 0.04445235803723335, 0.013818440027534962, -0.03091418743133545, 0.07037099450826645, 0.00656122388318181, -0.02051335759460926, 0.011605988256633282, -0.0040564266964793205, 0.046770330518484116, -0.009697603061795235, 0.022402862086892128, 0.04765422...