Unnamed: 0.1
int64
0
41k
Unnamed: 0
int64
0
41k
author
stringlengths
9
1.39k
id
stringlengths
11
18
summary
stringlengths
25
3.66k
title
stringlengths
4
258
year
int64
1.99k
2.02k
arxiv_url
stringlengths
32
39
info
stringlengths
523
3.18k
embeddings
stringlengths
16.9k
17.1k
1,300
1,300
['Karol Gregor', 'Ivo Danihelka', 'Alex Graves', 'Danilo Jimenez Rezende', 'Daan Wierstra']
1502.04623v2
This paper introduces the Deep Recurrent Attentive Writer (DRAW) neural network architecture for image generation. DRAW networks combine a novel spatial attention mechanism that mimics the foveation of the human eye, with a sequential variational auto-encoding framework that allows for the iterative construction of com...
DRAW: A Recurrent Neural Network For Image Generation
2,015
http://arxiv.org/pdf/1502.04623v2
Title DRAW Recurrent Neural Network Image Generation Summary paper introduces Deep Recurrent Attentive Writer DRAW neural network architecture image generation DRAW network combine novel spatial attention mechanism mimic foveation human eye sequential variational autoencoding framework allows iterative construction com...
[0.0017783779185265303, 0.04109233617782593, -0.017750965431332588, 0.031905073672533035, -0.0052001383155584335, -0.02110379934310913, 0.04627026617527008, -0.04339786246418953, -0.02106143720448017, 0.014042980968952179, 0.036026954650878906, -0.010988812893629074, 0.03505922853946686, 0.06298777461051941, 0.03344446...
1,301
1,301
['Zhiyong Cheng', 'Daniel Soudry', 'Zexi Mao', 'Zhenzhong Lan']
1503.03562v3
Compared to Multilayer Neural Networks with real weights, Binary Multilayer Neural Networks (BMNNs) can be implemented more efficiently on dedicated hardware. BMNNs have been demonstrated to be effective on binary classification tasks with Expectation BackPropagation (EBP) algorithm on high dimensional text datasets. I...
Training Binary Multilayer Neural Networks for Image Classification using Expectation Backpropagation
2,015
http://arxiv.org/pdf/1503.03562v3
Title Training Binary Multilayer Neural Networks Image Classification using Expectation Backpropagation Summary Compared Multilayer Neural Networks real weight Binary Multilayer Neural Networks BMNNs implemented efficiently dedicated hardware BMNNs demonstrated effective binary classification task Expectation BackPropa...
[0.024058552458882332, 0.008432784117758274, -0.02620229497551918, 0.04989316314458847, -0.0011386690894141793, 0.003531139576807618, 0.06868599355220795, -0.008158165961503983, 0.044750895351171494, -6.13533120485954e-05, 0.006127421278506517, 9.424835297977552e-05, -0.0002391389134572819, 0.03423604369163513, 0.01930...
1,302
1,302
['Yoojin Choi', 'Mostafa El-Khamy', 'Jungwon Lee']
1612.01543v2
Network quantization is one of network compression techniques to reduce the redundancy of deep neural networks. It reduces the number of distinct network parameter values by quantization in order to save the storage for them. In this paper, we design network quantization schemes that minimize the performance loss due t...
Towards the Limit of Network Quantization
2,016
http://arxiv.org/pdf/1612.01543v2
Title Towards Limit Network Quantization Summary Network quantization one network compression technique reduce redundancy deep neural network reduces number distinct network parameter value quantization order save storage paper design network quantization scheme minimize performance loss due quantization given compress...
[-0.06960011273622513, 0.032032303512096405, -0.022486310452222824, 0.03672216460108757, -0.0029707239009439945, -0.012439293786883354, 0.030588321387767792, 0.04969857633113861, -0.028085561469197273, 0.02317476086318493, 0.042100828140974045, 0.01901458390057087, -0.012496575713157654, 0.01985854282975197, 0.04748814...
1,303
1,303
['Shervin Ardeshir', 'Krishna Regmi', 'Ali Borji']
1612.05836v1
Mirror neurons have been observed in the primary motor cortex of primate species, in particular in humans and monkeys. A mirror neuron fires when a person performs a certain action, and also when he observes the same action being performed by another person. A crucial step towards building fully autonomous intelligent ...
EgoTransfer: Transferring Motion Across Egocentric and Exocentric Domains using Deep Neural Networks
2,016
http://arxiv.org/pdf/1612.05836v1
Title EgoTransfer Transferring Motion Across Egocentric Exocentric Domains using Deep Neural Networks Summary Mirror neuron observed primary motor cortex primate specie particular human monkey mirror neuron fire person performs certain action also observes action performed another person crucial step towards building f...
[-0.004367218818515539, 0.034670110791921616, -0.023826781660318375, 0.014749806374311447, -0.012711267918348312, 0.01432008109986782, 0.02313113398849964, 0.029733125120401382, -0.01558218989521265, -0.05070984363555908, -0.06232583895325661, -0.027369622141122818, 0.028064845129847527, 0.02280389703810215, 0.04780860...
1,304
1,304
['Ashish Shrivastava', 'Tomas Pfister', 'Oncel Tuzel', 'Josh Susskind', 'Wenda Wang', 'Russ Webb']
1612.07828v2
With recent progress in graphics, it has become more tractable to train models on synthetic images, potentially avoiding the need for expensive annotations. However, learning from synthetic images may not achieve the desired performance due to a gap between synthetic and real image distributions. To reduce this gap, we...
Learning from Simulated and Unsupervised Images through Adversarial Training
2,016
http://arxiv.org/pdf/1612.07828v2
Title Learning Simulated Unsupervised Images Adversarial Training Summary recent progress graphic become tractable train model synthetic image potentially avoiding need expensive annotation However learning synthetic image may achieve desired performance due gap synthetic real image distribution reduce gap propose Simu...
[0.008452516980469227, 0.035691242665052414, -0.007215404883027077, 0.03475160151720047, 0.03605329245328903, 0.0006339570973068476, 0.04135502129793167, -0.025885725393891335, 0.04193589463829994, 0.001971232006326318, -0.03894110023975372, 0.03422105684876442, -0.02724250592291355, 0.0243313480168581, 0.0659190267324...
1,305
1,305
['Joseph Shtok', 'Michael Zibulevsky', 'Michael Elad']
1311.7251v1
We propose a supervised machine learning approach for boosting existing signal and image recovery methods and demonstrate its efficacy on example of image reconstruction in computed tomography. Our technique is based on a local nonlinear fusion of several image estimates, all obtained by applying a chosen reconstructio...
Spatially-Adaptive Reconstruction in Computed Tomography using Neural Networks
2,013
http://arxiv.org/pdf/1311.7251v1
Title SpatiallyAdaptive Reconstruction Computed Tomography using Neural Networks Summary propose supervised machine learning approach boosting existing signal image recovery method demonstrate efficacy example image reconstruction computed tomography technique based local nonlinear fusion several image estimate obtaine...
[-0.02020111121237278, 0.07659915834665298, -0.017843853682279587, 0.0014149087946861982, -0.013125945813953876, 0.012646099552512169, 0.008067119866609573, 0.028104659169912338, 0.010141679085791111, 0.05731674283742905, -0.00208096276037395, 0.01821151003241539, 0.039573267102241516, 0.031352438032627106, 0.037561044...
1,306
1,306
['Min Lin', 'Qiang Chen', 'Shuicheng Yan']
1312.4400v3
We propose a novel deep network structure called "Network In Network" (NIN) to enhance model discriminability for local patches within the receptive field. The conventional convolutional layer uses linear filters followed by a nonlinear activation function to scan the input. Instead, we build micro neural networks with...
Network In Network
2,013
http://arxiv.org/pdf/1312.4400v3
Title Network Network Summary propose novel deep network structure called Network Network NIN enhance model discriminability local patch within receptive field conventional convolutional layer us linear filter followed nonlinear activation function scan input Instead build micro neural network complex structure abstrac...
[-0.03336009755730629, -0.04605528712272644, -0.03514127433300018, 0.06304863840341568, 0.029463447630405426, 0.0010377801954746246, 0.06113330274820328, -0.032949551939964294, -0.0012051536468788981, 0.044445812702178955, -0.020084911957383156, 0.007813499309122562, 0.02037225291132927, 0.04271400347352028, 0.03006796...
1,307
1,307
['Vu Pham', 'Théodore Bluche', 'Christopher Kermorvant', 'Jérôme Louradour']
1312.4569v2
Recurrent neural networks (RNNs) with Long Short-Term memory cells currently hold the best known results in unconstrained handwriting recognition. We show that their performance can be greatly improved using dropout - a recently proposed regularization method for deep architectures. While previous works showed that dro...
Dropout improves Recurrent Neural Networks for Handwriting Recognition
2,013
http://arxiv.org/pdf/1312.4569v2
Title Dropout improves Recurrent Neural Networks Handwriting Recognition Summary Recurrent neural network RNNs Long ShortTerm memory cell currently hold best known result unconstrained handwriting recognition show performance greatly improved using dropout recently proposed regularization method deep architecture previ...
[0.008852856233716011, 0.051547881215810776, 0.0005029726889915764, 0.08745897561311722, -0.02823861688375473, 0.015369447879493237, 0.06136412173509598, 0.013893031515181065, 0.01685958169400692, -0.008262834511697292, -0.030648333951830864, -0.03565920516848564, 0.03755892813205719, 0.03737610578536987, -0.0106246713...
1,308
1,308
['Yunlong He', 'Koray Kavukcuoglu', 'Yun Wang', 'Arthur Szlam', 'Yanjun Qi']
1312.5783v1
In this paper, we propose a new unsupervised feature learning framework, namely Deep Sparse Coding (DeepSC), that extends sparse coding to a multi-layer architecture for visual object recognition tasks. The main innovation of the framework is that it connects the sparse-encoders from different layers by a sparse-to-den...
Unsupervised Feature Learning by Deep Sparse Coding
2,013
http://arxiv.org/pdf/1312.5783v1
Title Unsupervised Feature Learning Deep Sparse Coding Summary paper propose new unsupervised feature learning framework namely Deep Sparse Coding DeepSC extends sparse coding multilayer architecture visual object recognition task main innovation framework connects sparseencoders different layer sparsetodense module sp...
[-0.044519152492284775, 0.020738838240504265, 0.015150311402976513, 0.03925582021474838, 0.004801414906978607, 0.03000466898083687, 0.044449012726545334, -0.014792164787650108, -0.002006933558732271, 0.016832465305924416, 0.022060632705688477, 0.0530753917992115, -0.0037354000378400087, 0.06767348945140839, 0.020597558...
1,309
1,309
['Michael Mathieu', 'Mikael Henaff', 'Yann LeCun']
1312.5851v5
Convolutional networks are one of the most widely employed architectures in computer vision and machine learning. In order to leverage their ability to learn complex functions, large amounts of data are required for training. Training a large convolutional network to produce state-of-the-art results can take weeks, eve...
Fast Training of Convolutional Networks through FFTs
2,013
http://arxiv.org/pdf/1312.5851v5
Title Fast Training Convolutional Networks FFTs Summary Convolutional network one widely employed architecture computer vision machine learning order leverage ability learn complex function large amount data required training Training large convolutional network produce stateoftheart result take week even using modern ...
[-0.024386318400502205, 0.026609787717461586, -0.009662763215601444, 0.0774223655462265, 0.001001889817416668, -0.0019500593189150095, 0.03364475443959236, 0.014185991138219833, 0.00015855419042054564, 0.0368589349091053, -0.01829111948609352, -0.005049149505794048, 0.031748753041028976, 0.0386483259499073, 0.016550621...
1,310
1,310
['Mohammad Ali Keyvanrad', 'Mohammad Pezeshki', 'Mohammad Ali Homayounpour']
1312.6158v2
Deep Belief Networks which are hierarchical generative models are effective tools for feature representation and extraction. Furthermore, DBNs can be used in numerous aspects of Machine Learning such as image denoising. In this paper, we propose a novel method for image denoising which relies on the DBNs' ability in fe...
Deep Belief Networks for Image Denoising
2,013
http://arxiv.org/pdf/1312.6158v2
Title Deep Belief Networks Image Denoising Summary Deep Belief Networks hierarchical generative model effective tool feature representation extraction Furthermore DBNs used numerous aspect Machine Learning image denoising paper propose novel method image denoising relies DBNs ability feature representation work based u...
[-0.030769752338528633, 0.0702359527349472, 0.003985796123743057, 0.05939345061779022, 1.0449864021211397e-05, 0.02481059916317463, 0.019002031534910202, 0.02489212155342102, -0.005435470957309008, 0.04178979992866516, -0.009952136315405369, 0.004764100071042776, -0.0032625223975628614, 0.014003160409629345, 0.00134790...
1,311
1,311
['Thomas Paine', 'Hailin Jin', 'Jianchao Yang', 'Zhe Lin', 'Thomas Huang']
1312.6186v1
The ability to train large-scale neural networks has resulted in state-of-the-art performance in many areas of computer vision. These results have largely come from computational break throughs of two forms: model parallelism, e.g. GPU accelerated training, which has seen quick adoption in computer vision circles, and ...
GPU Asynchronous Stochastic Gradient Descent to Speed Up Neural Network Training
2,013
http://arxiv.org/pdf/1312.6186v1
Title GPU Asynchronous Stochastic Gradient Descent Speed Neural Network Training Summary ability train largescale neural network resulted stateoftheart performance many area computer vision result largely come computational break throughs two form model parallelism eg GPU accelerated training seen quick adoption comput...
[-0.017272833734750748, 0.03319893032312393, -0.01644970290362835, 0.04191631078720093, -0.015076985582709312, -0.0016453847056254745, 0.07198367267847061, -0.012155565433204174, -0.016516122967004776, 0.004625339061021805, 0.0384073369204998, 0.0008678195299580693, 0.030106838792562485, 0.044469837099313736, 0.0237324...
1,312
1,312
['Christian Szegedy', 'Wojciech Zaremba', 'Ilya Sutskever', 'Joan Bruna', 'Dumitru Erhan', 'Ian Goodfellow', 'Rob Fergus']
1312.6199v4
Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we ...
Intriguing properties of neural networks
2,013
http://arxiv.org/pdf/1312.6199v4
Title Intriguing property neural network Summary Deep neural network highly expressive model recently achieved state art performance speech visual recognition task expressiveness reason succeed also cause learn uninterpretable solution could counterintuitive property paper report two property First find distinction ind...
[-0.008845947682857513, 0.057704370468854904, -0.006813005544245243, 0.03917571157217026, -0.01755141280591488, -0.007132187485694885, 0.03493347764015198, -0.0373954139649868, -0.0638759508728981, 0.006622873712331057, -0.0374855138361454, 0.017015593126416206, 0.015501551330089569, 0.08521293848752975, 0.034988105297...
1,313
1,313
['Joan Bruna', 'Wojciech Zaremba', 'Arthur Szlam', 'Yann LeCun']
1312.6203v3
Convolutional Neural Networks are extremely efficient architectures in image and audio recognition tasks, thanks to their ability to exploit the local translational invariance of signal classes over their domain. In this paper we consider possible generalizations of CNNs to signals defined on more general domains witho...
Spectral Networks and Locally Connected Networks on Graphs
2,013
http://arxiv.org/pdf/1312.6203v3
Title Spectral Networks Locally Connected Networks Graphs Summary Convolutional Neural Networks extremely efficient architecture image audio recognition task thanks ability exploit local translational invariance signal class domain paper consider possible generalization CNNs signal defined general domain without action...
[0.008198616094887257, -0.0042268685065209866, -0.006278360728174448, 0.09561203420162201, 0.006066532339900732, -0.03312276303768158, 0.05739676579833031, -0.015439274720847607, -0.0027652857825160027, 0.04089294373989105, -0.08994144946336746, 0.05151617154479027, -0.001836264505982399, 0.037779610604047775, 0.052496...
1,314
1,314
['Judy Hoffman', 'Eric Tzeng', 'Jeff Donahue', 'Yangqing Jia', 'Kate Saenko', 'Trevor Darrell']
1312.6204v2
Dataset bias remains a significant barrier towards solving real world computer vision tasks. Though deep convolutional networks have proven to be a competitive approach for image classification, a question remains: have these models have solved the dataset bias problem? In general, training or fine-tuning a state-of-th...
One-Shot Adaptation of Supervised Deep Convolutional Models
2,013
http://arxiv.org/pdf/1312.6204v2
Title OneShot Adaptation Supervised Deep Convolutional Models Summary Dataset bias remains significant barrier towards solving real world computer vision task Though deep convolutional network proven competitive approach image classification question remains model solved dataset bias problem general training finetuning...
[0.026009783148765564, 0.08111614733934402, -0.045873697847127914, 0.044434599578380585, 0.012719114311039448, 0.01934979483485222, 0.07865329831838608, 0.02224268764257431, -0.017189238220453262, -0.008699648082256317, -0.036776892840862274, 0.046881284564733505, -0.04224628210067749, 0.02610284835100174, 0.0055381599...
1,315
1,315
['Brody Huval', 'Adam Coates', 'Andrew Ng']
1312.6885v1
We investigate the use of deep neural networks for the novel task of class generic object detection. We show that neural networks originally designed for image recognition can be trained to detect objects within images, regardless of their class, including objects for which no bounding box labels have been provided. In...
Deep learning for class-generic object detection
2,013
http://arxiv.org/pdf/1312.6885v1
Title Deep learning classgeneric object detection Summary investigate use deep neural network novel task class generic object detection show neural network originally designed image recognition trained detect object within image regardless class including object bounding box label provided addition show bounding box la...
[-0.008585963398218155, 0.006269516889005899, 0.0018572964472696185, 0.09971366822719574, 0.016242891550064087, -0.020286506041884422, 0.0340019091963768, 0.011823536828160286, -0.011319578625261784, -0.006131285801529884, -0.00014983791334088892, 0.03624932840466499, -0.018365537747740746, 0.04912709817290306, -0.0018...
1,316
1,316
['Marijn Stollenga', 'Jonathan Masci', 'Faustino Gomez', 'Juergen Schmidhuber']
1407.3068v2
Traditional convolutional neural networks (CNN) are stationary and feedforward. They neither change their parameters during evaluation nor use feedback from higher to lower layers. Real brains, however, do. So does our Deep Attention Selective Network (dasNet) architecture. DasNets feedback structure can dynamically al...
Deep Networks with Internal Selective Attention through Feedback Connections
2,014
http://arxiv.org/pdf/1407.3068v2
Title Deep Networks Internal Selective Attention Feedback Connections Summary Traditional convolutional neural network CNN stationary feedforward neither change parameter evaluation use feedback higher lower layer Real brain however Deep Attention Selective Network dasNet architecture DasNets feedback structure dynamic...
[-0.014322888106107712, 0.01122693344950676, -0.02709897607564926, 0.032704103738069534, 0.029871849343180656, -0.031038913875818253, 0.030818011611700058, -0.009963291697204113, -0.012478266842663288, 0.02102833427488804, -0.03348695486783981, 0.0018134694546461105, 0.012514743953943253, 0.02509378083050251, 0.0471220...
1,317
1,317
['Ken Chatfield', 'Karen Simonyan', 'Andrew Zisserman']
1407.4764v3
We investigate the gains in precision and speed, that can be obtained by using Convolutional Networks (ConvNets) for on-the-fly retrieval - where classifiers are learnt at run time for a textual query from downloaded images, and used to rank large image or video datasets. We make three contributions: (i) we present a...
Efficient On-the-fly Category Retrieval using ConvNets and GPUs
2,014
http://arxiv.org/pdf/1407.4764v3
Title Efficient Onthefly Category Retrieval using ConvNets GPUs Summary investigate gain precision speed obtained using Convolutional Networks ConvNets onthefly retrieval classifier learnt run time textual query downloaded image used rank large image video datasets make three contribution present evaluation stateofthea...
[0.03286059945821762, -0.0010234956862404943, -0.017027413472533226, 0.02666580304503441, -0.004765307065099478, -0.006548562087118626, 0.006739594042301178, 0.03600779175758362, -0.06883855909109116, -0.0530836284160614, -0.013522394932806492, 0.006351022515445948, 0.004744537640362978, 0.01664545387029648, -0.0031238...
1,318
1,318
['Jure Žbontar', 'Yann LeCun']
1409.4326v2
We present a method for extracting depth information from a rectified image pair. We train a convolutional neural network to predict how well two image patches match and use it to compute the stereo matching cost. The cost is refined by cross-based cost aggregation and semiglobal matching, followed by a left-right cons...
Computing the Stereo Matching Cost with a Convolutional Neural Network
2,014
http://arxiv.org/pdf/1409.4326v2
Title Computing Stereo Matching Cost Convolutional Neural Network Summary present method extracting depth information rectified image pair train convolutional neural network predict well two image patch match use compute stereo matching cost cost refined crossbased cost aggregation semiglobal matching followed leftrigh...
[-0.030068855732679367, 0.0582142174243927, 0.03425713628530502, 0.10094945132732391, -0.03638007491827011, 0.0003419566201046109, -0.00929180160164833, -0.00936177745461464, -0.030891478061676025, -0.01390739344060421, -0.0034376110415905714, 0.021006803959608078, 0.06455032527446747, 0.06482187658548355, 0.0043764756...
1,319
1,319
['Zhicheng Yan', 'Hao Zhang', 'Robinson Piramuthu', 'Vignesh Jagadeesh', 'Dennis DeCoste', 'Wei Di', 'Yizhou Yu']
1410.0736v4
In image classification, visual separability between different object categories is highly uneven, and some categories are more difficult to distinguish than others. Such difficult categories demand more dedicated classifiers. However, existing deep convolutional neural networks (CNN) are trained as flat N-way classifi...
HD-CNN: Hierarchical Deep Convolutional Neural Network for Large Scale Visual Recognition
2,014
http://arxiv.org/pdf/1410.0736v4
Title HDCNN Hierarchical Deep Convolutional Neural Network Large Scale Visual Recognition Summary image classification visual separability different object category highly uneven category difficult distinguish others difficult category demand dedicated classifier However existing deep convolutional neural network CNN t...
[0.022387385368347168, 0.05958885699510574, -0.006671683397144079, 0.08576570451259613, -0.0019180984236299992, 0.018935950472950935, 0.06039523705840111, 0.00426028948277235, -0.015519977547228336, -0.022600878030061722, -0.011953962035477161, 0.017006825655698776, -0.03956787288188934, 0.004457027651369572, -0.008913...
1,320
1,320
['Tao Chen', 'Damian Borth', 'Trevor Darrell', 'Shih-Fu Chang']
1410.8586v1
This paper introduces a visual sentiment concept classification method based on deep convolutional neural networks (CNNs). The visual sentiment concepts are adjective noun pairs (ANPs) automatically discovered from the tags of web photos, and can be utilized as effective statistical cues for detecting emotions depicted...
DeepSentiBank: Visual Sentiment Concept Classification with Deep Convolutional Neural Networks
2,014
http://arxiv.org/pdf/1410.8586v1
Title DeepSentiBank Visual Sentiment Concept Classification Deep Convolutional Neural Networks Summary paper introduces visual sentiment concept classification method based deep convolutional neural network CNNs visual sentiment concept adjective noun pair ANPs automatically discovered tag web photo utilized effective ...
[0.030261268839240074, 0.03604789450764656, -0.014705969020724297, 0.08908987045288086, 0.005400666035711765, 0.029517762362957, -0.010843643918633461, 0.020021528005599976, 0.012162886559963226, -0.04995451495051384, -0.051597047597169876, 0.005730962846428156, -0.028457680717110634, 0.07399749010801315, -0.0165408886...
1,321
1,321
['Jonathan Long', 'Ning Zhang', 'Trevor Darrell']
1411.1091v1
Convolutional neural nets (convnets) trained from massive labeled datasets have substantially improved the state-of-the-art in image classification and object detection. However, visual understanding requires establishing correspondence on a finer level than object category. Given their large pooling regions and traini...
Do Convnets Learn Correspondence?
2,014
http://arxiv.org/pdf/1411.1091v1
Title Convnets Learn Correspondence Summary Convolutional neural net convnets trained massive labeled datasets substantially improved stateoftheart image classification object detection However visual understanding requires establishing correspondence finer level object category Given large pooling region training whol...
[0.023133477196097374, 0.03327110409736633, 0.012208646163344383, 0.07124073058366776, -0.01962212659418583, -0.02663523145020008, 0.03229932487010956, -0.0026548102032393217, -0.042134303599596024, -0.01125897653400898, -0.04777734726667404, 0.027497997507452965, 0.023928327485919, 0.01340468693524599, -0.006731848698...
1,322
1,322
['Zetao Chen', 'Obadiah Lam', 'Adam Jacobson', 'Michael Milford']
1411.1509v1
Recently Convolutional Neural Networks (CNNs) have been shown to achieve state-of-the-art performance on various classification tasks. In this paper, we present for the first time a place recognition technique based on CNN models, by combining the powerful features learnt by CNNs with a spatial and sequential filter. A...
Convolutional Neural Network-based Place Recognition
2,014
http://arxiv.org/pdf/1411.1509v1
Title Convolutional Neural Networkbased Place Recognition Summary Recently Convolutional Neural Networks CNNs shown achieve stateoftheart performance various classification task paper present first time place recognition technique based CNN model combining powerful feature learnt CNNs spatial sequential filter Applying...
[0.0004130738088861108, -0.001455445890314877, 0.01143156923353672, 0.026685068383812904, -0.007300737779587507, 0.015004602260887623, 0.052315838634967804, -0.030506029725074768, 0.007482957094907761, -0.008533368818461895, -0.003913660533726215, 0.04279301315546036, 0.020630721002817154, 0.012364625930786133, 0.07606...
1,323
1,323
['Mingmin Zhao', 'Chengxu Zhuang', 'Yizhou Wang', 'Tai Sing Lee']
1411.3815v6
We propose a new neurally-inspired model that can learn to encode the global relationship context of visual events across time and space and to use the contextual information to modulate the analysis by synthesis process in a predictive coding framework. The model learns latent contextual representations by maximizing ...
Predictive Encoding of Contextual Relationships for Perceptual Inference, Interpolation and Prediction
2,014
http://arxiv.org/pdf/1411.3815v6
Title Predictive Encoding Contextual Relationships Perceptual Inference Interpolation Prediction Summary propose new neurallyinspired model learn encode global relationship context visual event across time space use contextual information modulate analysis synthesis process predictive coding framework model learns late...
[0.02127302810549736, 0.04086500406265259, -0.0160154327750206, 0.0432288758456707, -0.011826969683170319, -0.006210337858647108, 0.024612175300717354, 0.021976230666041374, -0.032276153564453125, 0.003017231822013855, 0.01629154570400715, -0.029293568804860115, 0.024843405932188034, 0.09069018065929413, 0.017777539789...
1,324
1,324
['Karel Lenc', 'Andrea Vedaldi']
1411.5908v2
Despite the importance of image representations such as histograms of oriented gradients and deep Convolutional Neural Networks (CNN), our theoretical understanding of them remains limited. Aiming at filling this gap, we investigate three key mathematical properties of representations: equivariance, invariance, and equ...
Understanding image representations by measuring their equivariance and equivalence
2,014
http://arxiv.org/pdf/1411.5908v2
Title Understanding image representation measuring equivariance equivalence Summary Despite importance image representation histogram oriented gradient deep Convolutional Neural Networks CNN theoretical understanding remains limited Aiming filling gap investigate three key mathematical property representation equivaria...
[-0.02513744868338108, -0.004452292341738939, -0.010958888567984104, 0.02803938090801239, -0.02350863628089428, 0.0017106188461184502, 0.03982926532626152, 0.01014879159629345, -0.07799512892961502, -0.011668948456645012, -0.02106003649532795, 0.040924202650785446, 0.007518621627241373, 0.016990380361676216, 0.03466475...
1,325
1,325
['Alexey Dosovitskiy', 'Jost Tobias Springenberg', 'Maxim Tatarchenko', 'Thomas Brox']
1411.5928v4
We train generative 'up-convolutional' neural networks which are able to generate images of objects given object style, viewpoint, and color. We train the networks on rendered 3D models of chairs, tables, and cars. Our experiments show that the networks do not merely learn all images by heart, but rather find a meaning...
Learning to Generate Chairs, Tables and Cars with Convolutional Networks
2,014
http://arxiv.org/pdf/1411.5928v4
Title Learning Generate Chairs Tables Cars Convolutional Networks Summary train generative upconvolutional neural network able generate image object given object style viewpoint color train network rendered 3D model chair table car experiment show network merely learn image heart rather find meaningful representation 3...
[-0.006186934188008308, 0.08486000448465347, 0.017374316230416298, 0.07650690525770187, -0.0015747826546430588, 0.015066166408360004, 0.006135623436421156, 0.007469245232641697, -0.05414315685629845, -0.008864901028573513, -0.025640543550252914, 0.034234825521707535, -0.014441086910665035, 0.041411980986595154, 0.05579...
1,326
1,326
['Donggeun Yoo', 'Sunggyun Park', 'Joon-Young Lee', 'In So Kweon']
1412.1628v2
Compared to image representation based on low-level local descriptors, deep neural activations of Convolutional Neural Networks (CNNs) are richer in mid-level representation, but poorer in geometric invariance properties. In this paper, we present a straightforward framework for better image representation by combining...
Fisher Kernel for Deep Neural Activations
2,014
http://arxiv.org/pdf/1412.1628v2
Title Fisher Kernel Deep Neural Activations Summary Compared image representation based lowlevel local descriptor deep neural activation Convolutional Neural Networks CNNs richer midlevel representation poorer geometric invariance property paper present straightforward framework better image representation combining tw...
[-0.05076296254992485, -0.004704094026237726, -0.01816350221633911, 0.09496255964040756, 0.006220074370503426, 0.008591861464083195, 0.03816264495253563, 0.005253775045275688, 0.013836436904966831, -0.028025880455970764, -0.02979557402431965, 0.047899842262268066, -0.027667861431837082, 0.05360148102045059, 0.034638911...
1,327
1,327
['Soren Goyal', 'Paul Benjamin']
1412.3684v1
Recognition of objects using Deep Neural Networks is an active area of research and many breakthroughs have been made in the last few years. The paper attempts to indicate how far this field has progressed. The paper briefly describes the history of research in Neural Networks and describe several of the recent advance...
Object Recognition Using Deep Neural Networks: A Survey
2,014
http://arxiv.org/pdf/1412.3684v1
Title Object Recognition Using Deep Neural Networks Survey Summary Recognition object using Deep Neural Networks active area research many breakthrough made last year paper attempt indicate far field progressed paper briefly describes history research Neural Networks describe several recent advance field performance re...
[0.009895606897771358, 0.05109161138534546, 0.033230606466531754, 0.10092389583587646, -0.030334502458572388, 0.020600909367203712, 0.0549166202545166, -0.011171186342835426, -0.005730865523219109, 0.015467175282537937, -0.02198457531630993, 0.019018935039639473, 0.011872165836393833, 0.0011352202855050564, -0.01635651...
1,328
1,328
['Shixiang Gu', 'Luca Rigazio']
1412.5068v4
Recent work has shown deep neural networks (DNNs) to be highly susceptible to well-designed, small perturbations at the input layer, or so-called adversarial examples. Taking images as an example, such distortions are often imperceptible, but can result in 100% mis-classification for a state of the art DNN. We study th...
Towards Deep Neural Network Architectures Robust to Adversarial Examples
2,014
http://arxiv.org/pdf/1412.5068v4
Title Towards Deep Neural Network Architectures Robust Adversarial Examples Summary Recent work shown deep neural network DNNs highly susceptible welldesigned small perturbation input layer socalled adversarial example Taking image example distortion often imperceptible result 100 misclassification state art DNN study ...
[0.02023523859679699, 0.04792892932891846, -0.04135575145483017, 0.030075907707214355, -0.00019697290554177016, -0.01254831999540329, 0.025744983926415443, -0.019171632826328278, 0.012688565999269485, -0.009471794590353966, 0.01662198267877102, 0.004160782787948847, -0.0003272203612141311, 0.05058758333325386, 0.074368...
1,329
1,329
['Angjoo Kanazawa', 'Abhishek Sharma', 'David Jacobs']
1412.5104v1
Convolutional Neural Networks (ConvNets) have shown excellent results on many visual classification tasks. With the exception of ImageNet, these datasets are carefully crafted such that objects are well-aligned at similar scales. Naturally, the feature learning problem gets more challenging as the amount of variation i...
Locally Scale-Invariant Convolutional Neural Networks
2,014
http://arxiv.org/pdf/1412.5104v1
Title Locally ScaleInvariant Convolutional Neural Networks Summary Convolutional Neural Networks ConvNets shown excellent result many visual classification task exception ImageNet datasets carefully crafted object wellaligned similar scale Naturally feature learning problem get challenging amount variation data increas...
[-0.0031542659271508455, 0.029367204755544662, -0.023211918771266937, 0.06599003821611404, 0.029030373319983482, 0.0005972482031211257, 0.06317582726478577, 0.000591798743698746, -0.023247210308909416, 0.03897477313876152, -0.0372278057038784, 0.04436412826180458, -0.04080982878804207, 0.04918874427676201, 0.0365423150...
1,330
1,330
['Yunchao Gong', 'Liu Liu', 'Ming Yang', 'Lubomir Bourdev']
1412.6115v1
Deep convolutional neural networks (CNN) has become the most promising method for object recognition, repeatedly demonstrating record breaking results for image classification and object detection in recent years. However, a very deep CNN generally involves many layers with millions of parameters, making the storage of...
Compressing Deep Convolutional Networks using Vector Quantization
2,014
http://arxiv.org/pdf/1412.6115v1
Title Compressing Deep Convolutional Networks using Vector Quantization Summary Deep convolutional neural network CNN become promising method object recognition repeatedly demonstrating record breaking result image classification object detection recent year However deep CNN generally involves many layer million parame...
[-0.03430059924721718, 0.06364308297634125, -0.011925095692276955, 0.07501509040594101, 0.018139243125915527, 0.005381346680223942, 0.055984869599342346, 0.042527057230472565, -0.05742371082305908, 0.011542701162397861, 0.0016750154318287969, 0.03191405162215233, -0.0059597669169306755, 0.020212514325976372, 0.02050597...
1,331
1,331
['Jifeng Dai', 'Yang Lu', 'Ying-Nian Wu']
1412.6296v2
The convolutional neural networks (CNNs) have proven to be a powerful tool for discriminative learning. Recently researchers have also started to show interest in the generative aspects of CNNs in order to gain a deeper understanding of what they have learned and how to further improve them. This paper investigates gen...
Generative Modeling of Convolutional Neural Networks
2,014
http://arxiv.org/pdf/1412.6296v2
Title Generative Modeling Convolutional Neural Networks Summary convolutional neural network CNNs proven powerful tool discriminative learning Recently researcher also started show interest generative aspect CNNs order gain deeper understanding learned improve paper investigates generative modeling CNNs main contributi...
[0.014649791643023491, 0.0131193483248353, -0.03608517721295357, 0.04045404866337776, 0.00331891723908484, -0.01087824534624815, 0.0518806017935276, -0.005715991836041212, -0.029510000720620155, 0.015606483444571495, 0.007697335910052061, 0.03507612645626068, -0.050713129341602325, 0.006542638409882784, 0.0400983728468...
1,332
1,332
['Scott Reed', 'Honglak Lee', 'Dragomir Anguelov', 'Christian Szegedy', 'Dumitru Erhan', 'Andrew Rabinovich']
1412.6596v3
Current state-of-the-art deep learning systems for visual object recognition and detection use purely supervised training with regularization such as dropout to avoid overfitting. The performance depends critically on the amount of labeled examples, and in current practice the labels are assumed to be unambiguous and a...
Training Deep Neural Networks on Noisy Labels with Bootstrapping
2,014
http://arxiv.org/pdf/1412.6596v3
Title Training Deep Neural Networks Noisy Labels Bootstrapping Summary Current stateoftheart deep learning system visual object recognition detection use purely supervised training regularization dropout avoid overfitting performance depends critically amount labeled example current practice label assumed unambiguous a...
[0.03390777111053467, 0.06980360299348831, 5.445735587272793e-05, 0.04710793122649193, 0.03641525283455849, 0.026089133694767952, 0.03462924063205719, 0.003256925381720066, 0.007399866357445717, 0.008724432438611984, -0.0388440266251564, 0.017657335847616196, -0.017454469576478004, 0.038845695555210114, 0.0171876326203...
1,333
1,333
['Jyh-Jing Hwang', 'Tyng-Luh Liu']
1412.6857v5
We address the problem of contour detection via per-pixel classifications of edge point. To facilitate the process, the proposed approach leverages with DenseNet, an efficient implementation of multiscale convolutional neural networks (CNNs), to extract an informative feature vector for each pixel and uses an SVM class...
Contour Detection Using Cost-Sensitive Convolutional Neural Networks
2,014
http://arxiv.org/pdf/1412.6857v5
Title Contour Detection Using CostSensitive Convolutional Neural Networks Summary address problem contour detection via perpixel classification edge point facilitate process proposed approach leverage DenseNet efficient implementation multiscale convolutional neural network CNNs extract informative feature vector pixel...
[-0.04743803292512894, 0.018413642421364784, 0.00046210785512812436, 0.0913768857717514, -0.031704798340797424, -0.04441967234015465, 0.005486268550157547, 0.00991961918771267, -0.031080525368452072, 0.0233041662722826, 0.000132516332087107, 0.05918120592832565, -0.023787962272763252, 0.06802939623594284, -0.0348731428...
1,334
1,334
['Michael Giering', 'Vivek Venugopalan', 'Kishore Reddy']
1412.7006v2
The ability to simultaneously leverage multiple modes of sensor information is critical for perception of an automated vehicle's physical surroundings. Spatio-temporal alignment of registration of the incoming information is often a prerequisite to analyzing the fused data. The persistence and reliability of multi-moda...
Multi-modal Sensor Registration for Vehicle Perception via Deep Neural Networks
2,014
http://arxiv.org/pdf/1412.7006v2
Title Multimodal Sensor Registration Vehicle Perception via Deep Neural Networks Summary ability simultaneously leverage multiple mode sensor information critical perception automated vehicle physical surroundings Spatiotemporal alignment registration incoming information often prerequisite analyzing fused data persist...
[-0.03882865607738495, 0.0031739312689751387, 0.019992481917142868, 0.07435151189565659, -0.013046486303210258, 0.0063933259807527065, 0.032388318330049515, -0.057006292045116425, -0.036861203610897064, -0.01842789351940155, -0.009787047281861305, 0.0020383670926094055, 0.018613530322909355, 0.04887579381465912, 0.0297...
1,335
1,335
['Pierre Sermanet', 'Andrea Frome', 'Esteban Real']
1412.7054v3
This paper presents experiments extending the work of Ba et al. (2014) on recurrent neural models for attention into less constrained visual environments, specifically fine-grained categorization on the Stanford Dogs data set. In this work we use an RNN of the same structure but substitute a more powerful visual networ...
Attention for Fine-Grained Categorization
2,014
http://arxiv.org/pdf/1412.7054v3
Title Attention FineGrained Categorization Summary paper present experiment extending work Ba et al 2014 recurrent neural model attention le constrained visual environment specifically finegrained categorization Stanford Dogs data set work use RNN structure substitute powerful visual network perform largescale pretrain...
[0.01894681714475155, 0.02277916669845581, -0.03539706766605377, 0.042571771889925, 0.0397765152156353, 0.0251314640045166, 0.05948355048894882, -0.02350364625453949, -0.03916492685675621, -0.04635867848992348, -0.023934124037623405, 0.041696421802043915, 0.0008275941945612431, 0.027655083686113358, 0.05524893850088119...
1,336
1,336
['Deepak Pathak', 'Evan Shelhamer', 'Jonathan Long', 'Trevor Darrell']
1412.7144v4
Multiple instance learning (MIL) can reduce the need for costly annotation in tasks such as semantic segmentation by weakening the required degree of supervision. We propose a novel MIL formulation of multi-class semantic segmentation learning by a fully convolutional network. In this setting, we seek to learn a semant...
Fully Convolutional Multi-Class Multiple Instance Learning
2,014
http://arxiv.org/pdf/1412.7144v4
Title Fully Convolutional MultiClass Multiple Instance Learning Summary Multiple instance learning MIL reduce need costly annotation task semantic segmentation weakening required degree supervision propose novel MIL formulation multiclass semantic segmentation learning fully convolutional network setting seek learn sem...
[0.021199142560362816, 0.014690147712826729, 0.027312910184264183, 0.09554002434015274, 0.017748042941093445, 0.012972441501915455, 0.044761594384908676, -0.017923813313245773, -0.062246035784482956, -0.01158760767430067, -0.00704719964414835, 0.03602920100092888, -0.0581299290060997, -0.019356604665517807, 0.013802490...
1,337
1,337
['Francisco Massa', 'Mathieu Aubry', 'Renaud Marlet']
1412.7190v4
In this paper we study the application of convolutional neural networks for jointly detecting objects depicted in still images and estimating their 3D pose. We identify different feature representations of oriented objects, and energies that lead a network to learn this representations. The choice of the representation...
Convolutional Neural Networks for joint object detection and pose estimation: A comparative study
2,014
http://arxiv.org/pdf/1412.7190v4
Title Convolutional Neural Networks joint object detection pose estimation comparative study Summary paper study application convolutional neural network jointly detecting object depicted still image estimating 3D pose identify different feature representation oriented object energy lead network learn representation ch...
[-0.005348449107259512, 0.02463681809604168, 0.030960170552134514, 0.11003576219081879, 0.03491911664605141, 0.023098977282643318, 0.010167786851525307, -0.02550891786813736, -0.023507753387093544, -0.06678110361099243, -0.033693887293338776, 0.03281319886445999, 0.016345428302884102, 0.05199694260954857, 0.02106344886...
1,338
1,338
['Dong Wang', 'Xiaoyang Tan']
1412.7259v3
In this paper, we investigate the problem of learning feature representation from unlabeled data using a single-layer K-means network. A K-means network maps the input data into a feature representation by finding the nearest centroid for each input point, which has attracted researchers' great attention recently due t...
Unsupervised Feature Learning with C-SVDDNet
2,014
http://arxiv.org/pdf/1412.7259v3
Title Unsupervised Feature Learning CSVDDNet Summary paper investigate problem learning feature representation unlabeled data using singlelayer Kmeans network Kmeans network map input data feature representation finding nearest centroid input point attracted researcher great attention recently due simplicity effectiven...
[-0.057120613753795624, -0.06225677207112312, -0.008751084096729755, 0.03425849974155426, 0.027740564197301865, 0.0011171173537150025, 0.06252800673246384, -0.008105850778520107, 0.01870320364832878, 0.015367591753602028, -0.014805300161242485, 0.043712813407182693, -0.002790298545733094, 0.047680146992206573, 0.000114...
1,339
1,339
['Taco S. Cohen', 'Max Welling']
1412.7659v3
When a three-dimensional object moves relative to an observer, a change occurs on the observer's image plane and in the visual representation computed by a learned model. Starting with the idea that a good visual representation is one that transforms linearly under scene motions, we show, using the theory of group repr...
Transformation Properties of Learned Visual Representations
2,014
http://arxiv.org/pdf/1412.7659v3
Title Transformation Properties Learned Visual Representations Summary threedimensional object move relative observer change occurs observer image plane visual representation computed learned model Starting idea good visual representation one transforms linearly scene motion show using theory group representation repre...
[0.03304659575223923, 0.015819165855646133, 0.02191050536930561, 0.030849481001496315, -0.028198665007948875, 0.01489346008747816, 0.05208100378513336, 0.023314662277698517, -0.07428818196058273, -0.06409481167793274, -0.026510294526815414, 0.0791262611746788, 0.01112199667841196, 0.009424170479178429, 0.06010673567652...
1,340
1,340
['Jimmy Ba', 'Volodymyr Mnih', 'Koray Kavukcuoglu']
1412.7755v2
We present an attention-based model for recognizing multiple objects in images. The proposed model is a deep recurrent neural network trained with reinforcement learning to attend to the most relevant regions of the input image. We show that the model learns to both localize and recognize multiple objects despite being...
Multiple Object Recognition with Visual Attention
2,014
http://arxiv.org/pdf/1412.7755v2
Title Multiple Object Recognition Visual Attention Summary present attentionbased model recognizing multiple object image proposed model deep recurrent neural network trained reinforcement learning attend relevant region input image show model learns localize recognize multiple object despite given class label training...
[0.020174093544483185, -0.0028027223888784647, 0.027366604655981064, 0.05572930723428726, 0.021419784054160118, -0.006988062523305416, 0.05212864279747009, -0.016575373709201813, -0.010351050645112991, -0.05408230051398277, -0.02805120311677456, 0.008901647292077541, -0.0006918362923897803, 0.012942686676979065, 0.0258...
1,341
1,341
['Mark D. McDonnell', 'Migel D. Tissera', 'Tony Vladusich', 'André van Schaik', 'Jonathan Tapson']
1412.8307v2
Recent advances in training deep (multi-layer) architectures have inspired a renaissance in neural network use. For example, deep convolutional networks are becoming the default option for difficult tasks on large datasets, such as image and speech recognition. However, here we show that error rates below 1% on the MNI...
Fast, simple and accurate handwritten digit classification by training shallow neural network classifiers with the 'extreme learning machine' algorithm
2,014
http://arxiv.org/pdf/1412.8307v2
Title Fast simple accurate handwritten digit classification training shallow neural network classifier extreme learning machine algorithm Summary Recent advance training deep multilayer architecture inspired renaissance neural network use example deep convolutional network becoming default option difficult task large d...
[-0.034386683255434036, 0.005551040638238192, -0.021902387961745262, 0.05499494448304176, 0.030576156452298164, -0.025210225954651833, 0.037678271532058716, -0.002272348152473569, -0.005359406117349863, -0.0068536726757884026, -0.03171674534678459, 0.025404401123523712, 0.05267886072397232, 0.034963443875312805, 0.0175...
1,342
1,342
['Jun Li', 'Heyou Chang', 'Jian Yang']
1501.00777v1
Sparse coding can learn good robust representation to noise and model more higher-order representation for image classification. However, the inference algorithm is computationally expensive even though the supervised signals are used to learn compact and discriminative dictionaries in sparse coding techniques. Luckily...
Sparse Deep Stacking Network for Image Classification
2,015
http://arxiv.org/pdf/1501.00777v1
Title Sparse Deep Stacking Network Image Classification Summary Sparse coding learn good robust representation noise model higherorder representation image classification However inference algorithm computationally expensive even though supervised signal used learn compact discriminative dictionary sparse coding techni...
[-0.03585321828722954, 0.03874737769365311, -0.00037170780706219375, 0.06312096863985062, 0.03246019780635834, 0.019826369360089302, 0.04569540545344353, 0.013463765382766724, -0.01678251475095749, -0.010996973142027855, -0.026203619316220284, 0.03232604265213013, 0.01766858622431755, 0.030870119109749794, 0.0206135995...
1,343
1,343
['Tianyi Liu', 'Shuangsang Fang', 'Yuehui Zhao', 'Peng Wang', 'Jun Zhang']
1506.01195v2
Deep learning refers to the shining branch of machine learning that is based on learning levels of representations. Convolutional Neural Networks (CNN) is one kind of deep neural network. It can study concurrently. In this article, we gave a detailed analysis of the process of CNN algorithm both the forward process and...
Implementation of Training Convolutional Neural Networks
2,015
http://arxiv.org/pdf/1506.01195v2
Title Implementation Training Convolutional Neural Networks Summary Deep learning refers shining branch machine learning based learning level representation Convolutional Neural Networks CNN one kind deep neural network study concurrently article gave detailed analysis process CNN algorithm forward process back propaga...
[0.012884123250842094, -0.01853923127055168, -0.03982849791646004, 0.08761823177337646, -0.02721298672258854, 0.020875783637166023, 0.06596045941114426, 0.0024334066547453403, 0.003191978670656681, 0.000276155915344134, 0.015357938595116138, 0.027846544981002808, 0.02730819582939148, 0.051200203597545624, 0.03024302236...
1,344
1,344
['Alexey Dosovitskiy', 'Thomas Brox']
1506.02753v4
Feature representations, both hand-designed and learned ones, are often hard to analyze and interpret, even when they are extracted from visual data. We propose a new approach to study image representations by inverting them with an up-convolutional neural network. We apply the method to shallow representations (HOG, S...
Inverting Visual Representations with Convolutional Networks
2,015
http://arxiv.org/pdf/1506.02753v4
Title Inverting Visual Representations Convolutional Networks Summary Feature representation handdesigned learned one often hard analyze interpret even extracted visual data propose new approach study image representation inverting upconvolutional neural network apply method shallow representation HOG SIFT LBP well dee...
[-0.030265266075730324, 0.013498159125447273, -0.021164553239941597, 0.09882649034261703, 0.005388273391872644, -0.006625197362154722, -0.024673083797097206, 0.035175621509552, -0.04664892330765724, 0.009382136166095734, -0.052162639796733856, 0.03548150137066841, -0.006066895555704832, 0.02513040602207184, 0.025728357...
1,345
1,345
['Yiyi Liao', 'Sarath Kodagoda', 'Yue Wang', 'Lei Shi', 'Yong Liu']
1506.03899v1
Place classification is a fundamental ability that a robot should possess to carry out effective human-robot interactions. It is a nontrivial classification problem which has attracted many research. In recent years, there is a high exploitation of Artificial Intelligent algorithms in robotics applications. Inspired by...
Place classification with a graph regularized deep neural network model
2,015
http://arxiv.org/pdf/1506.03899v1
Title Place classification graph regularized deep neural network model Summary Place classification fundamental ability robot posse carry effective humanrobot interaction nontrivial classification problem attracted many research recent year high exploitation Artificial Intelligent algorithm robotics application Inspire...
[0.015440382994711399, 0.005156323779374361, -0.006396645680069923, -0.023342274129390717, -0.008342364802956581, -0.033554308116436005, 0.04852072149515152, -0.05316648259758949, 0.01343171764165163, -0.018610920757055283, -0.04503820091485977, 0.05659680441021919, 0.03273141384124756, 0.008172577247023582, 0.03308198...
1,346
1,346
['Mikael Henaff', 'Joan Bruna', 'Yann LeCun']
1506.05163v1
Deep Learning's recent successes have mostly relied on Convolutional Networks, which exploit fundamental statistical properties of images, sounds and video data: the local stationarity and multi-scale compositional structure, that allows expressing long range interactions in terms of shorter, localized interactions. Ho...
Deep Convolutional Networks on Graph-Structured Data
2,015
http://arxiv.org/pdf/1506.05163v1
Title Deep Convolutional Networks GraphStructured Data Summary Deep Learnings recent success mostly relied Convolutional Networks exploit fundamental statistical property image sound video data local stationarity multiscale compositional structure allows expressing long range interaction term shorter localized interact...
[0.00559247937053442, 0.05077776312828064, -0.009778850711882114, 0.03215586394071579, -0.01797165349125862, -0.03526365011930466, 0.05250825360417366, -0.0040105776861310005, -0.015482352115213871, 0.024518266320228577, -0.011448857374489307, 0.003774888813495636, 0.010075264610350132, 0.0879957303404808, 0.0377266444...
1,347
1,347
['Junting Pan', 'Xavier Giró-i-Nieto']
1507.01422v1
The prediction of saliency areas in images has been traditionally addressed with hand crafted features based on neuroscience principles. This paper however addresses the problem with a completely data-driven approach by training a convolutional network. The learning process is formulated as a minimization of a loss fun...
End-to-end Convolutional Network for Saliency Prediction
2,015
http://arxiv.org/pdf/1507.01422v1
Title Endtoend Convolutional Network Saliency Prediction Summary prediction saliency area image traditionally addressed hand crafted feature based neuroscience principle paper however address problem completely datadriven approach training convolutional network learning process formulated minimization loss function mea...
[-0.0053284442983567715, -0.012808958068490028, -0.0014501605182886124, 0.039911847561597824, 0.030681803822517395, -0.008064945228397846, 0.01310474518686533, 0.01804681122303009, -0.03463767096400261, -0.0050025321543216705, -0.04910210520029068, 0.03521096706390381, 0.011644710786640644, 0.010868269018828869, 0.0509...
1,348
1,348
['Joao Carreira', 'Pulkit Agrawal', 'Katerina Fragkiadaki', 'Jitendra Malik']
1507.06550v3
Hierarchical feature extractors such as Convolutional Networks (ConvNets) have achieved impressive performance on a variety of classification tasks using purely feedforward processing. Feedforward architectures can learn rich representations of the input space but do not explicitly model dependencies in the output spac...
Human Pose Estimation with Iterative Error Feedback
2,015
http://arxiv.org/pdf/1507.06550v3
Title Human Pose Estimation Iterative Error Feedback Summary Hierarchical feature extractor Convolutional Networks ConvNets achieved impressive performance variety classification task using purely feedforward processing Feedforward architecture learn rich representation input space explicitly model dependency output sp...
[0.009504173882305622, 0.039428289979696274, 0.006403577979654074, 0.06933015584945679, 0.033030539751052856, 0.039503879845142365, -0.0017843999667093158, 0.011922626756131649, 0.01681501232087612, -0.05483170598745346, -0.03928462788462639, -0.04203733801841736, 0.03649357333779335, 0.01997450925409794, 0.05206305906...
1,349
1,349
['Andreas Eitel', 'Jost Tobias Springenberg', 'Luciano Spinello', 'Martin Riedmiller', 'Wolfram Burgard']
1507.06821v2
Robust object recognition is a crucial ingredient of many, if not all, real-world robotics applications. This paper leverages recent progress on Convolutional Neural Networks (CNNs) and proposes a novel RGB-D architecture for object recognition. Our architecture is composed of two separate CNN processing streams - one ...
Multimodal Deep Learning for Robust RGB-D Object Recognition
2,015
http://arxiv.org/pdf/1507.06821v2
Title Multimodal Deep Learning Robust RGBD Object Recognition Summary Robust object recognition crucial ingredient many realworld robotics application paper leverage recent progress Convolutional Neural Networks CNNs proposes novel RGBD architecture object recognition architecture composed two separate CNN processing s...
[0.0267486572265625, 0.04654411971569061, 0.023529162630438805, 0.06812432408332825, -0.006076931953430176, -0.011071152985095978, 0.0031997847836464643, -0.06467542052268982, -0.06620202958583832, -0.051001716405153275, -0.032343462109565735, 0.03396422788500786, 0.0073140389285981655, 0.048996929079294205, 0.01432953...
1,350
1,350
['Zhanglin Peng', 'Ya Li', 'Zhaoquan Cai', 'Liang Lin']
1508.01887v2
This work investigates how the traditional image classification pipelines can be extended into a deep architecture, inspired by recent successes of deep neural networks. We propose a deep boosting framework based on layer-by-layer joint feature boosting and dictionary learning. In each layer, we construct a dictionary ...
Deep Boosting: Joint Feature Selection and Analysis Dictionary Learning in Hierarchy
2,015
http://arxiv.org/pdf/1508.01887v2
Title Deep Boosting Joint Feature Selection Analysis Dictionary Learning Hierarchy Summary work investigates traditional image classification pipeline extended deep architecture inspired recent success deep neural network propose deep boosting framework based layerbylayer joint feature boosting dictionary learning laye...
[-0.024400968104600906, 0.03256668895483017, -0.02135789953172207, 0.0426444336771965, 0.013378564268350601, 0.02457839623093605, -0.008432416245341301, 0.0012607467360794544, -0.004619675688445568, -0.022815698757767677, 0.0073996721766889095, 0.008286673575639725, 0.0018658209592103958, 0.06439267843961716, -0.006939...
1,351
1,351
['Mohammad Javad Shafiee', 'Parthipan Siva', 'Alexander Wong']
1508.05463v4
Deep neural networks is a branch in machine learning that has seen a meteoric rise in popularity due to its powerful abilities to represent and model high-level abstractions in highly complex data. One area in deep neural networks that is ripe for exploration is neural connectivity formation. A pivotal study on the bra...
StochasticNet: Forming Deep Neural Networks via Stochastic Connectivity
2,015
http://arxiv.org/pdf/1508.05463v4
Title StochasticNet Forming Deep Neural Networks via Stochastic Connectivity Summary Deep neural network branch machine learning seen meteoric rise popularity due powerful ability represent model highlevel abstraction highly complex data One area deep neural network ripe exploration neural connectivity formation pivota...
[-0.033324647694826126, 0.0343669131398201, -0.04456624388694763, 0.013093428686261177, 0.0367269366979599, -0.024621762335300446, 0.03201883286237717, -0.0643736720085144, -0.019769391044974327, 0.032747890800237656, -0.00020851512090303004, -0.02547348663210869, 0.04138711467385292, 0.07019155472517014, 0.04184658452...
1,352
1,352
['Patrick O. Glauner']
1508.06535v1
This thesis describes the design and implementation of a smile detector based on deep convolutional neural networks. It starts with a summary of neural networks, the difficulties of training them and new training methods, such as Restricted Boltzmann Machines or autoencoders. It then provides a literature review of con...
Deep Convolutional Neural Networks for Smile Recognition
2,015
http://arxiv.org/pdf/1508.06535v1
Title Deep Convolutional Neural Networks Smile Recognition Summary thesis describes design implementation smile detector based deep convolutional neural network start summary neural network difficulty training new training method Restricted Boltzmann Machines autoencoders provides literature review convolutional neural...
[0.02878011204302311, 0.058941759169101715, -0.013514582999050617, 0.04445911571383476, 0.04354594275355339, 0.011881659738719463, 0.03514307364821434, 0.02111666090786457, 0.05751587450504303, -0.008510320447385311, 0.02286430448293686, 0.011832254938781261, -0.012694060802459717, 0.04638998955488205, 0.05232420936226...
1,353
1,353
['Galin Georgiev']
1508.06585v5
We study from a physics viewpoint a class of generative neural nets, Gibbs machines, designed for gradual learning. While including variational auto-encoders, they offer a broader universal platform for incrementally adding newly learned features, including physical symmetries. Their direct connection to statistical ph...
Towards universal neural nets: Gibbs machines and ACE
2,015
http://arxiv.org/pdf/1508.06585v5
Title Towards universal neural net Gibbs machine ACE Summary study physic viewpoint class generative neural net Gibbs machine designed gradual learning including variational autoencoders offer broader universal platform incrementally adding newly learned feature including physical symmetry direct connection statistical...
[-0.028678923845291138, 0.04538720101118088, -0.010196555405855179, 0.008478876203298569, 0.000578559294808656, -0.005213339347392321, -0.000548098236322403, -0.016260843724012375, -0.026188772171735764, 0.02067079395055771, 0.017493456602096558, -0.0035618843976408243, 0.019489608705043793, 0.048160865902900696, 0.035...
1,354
1,354
['Markus Thom', 'Franz Gritschneder']
1508.06904v5
A rigorous formulation of the dynamics of a signal processing scheme aimed at dense signal scanning without any loss in accuracy is introduced and analyzed. Related methods proposed in the recent past lack a satisfactory analysis of whether they actually fulfill any exactness constraints. This is improved through an ex...
Rapid Exact Signal Scanning with Deep Convolutional Neural Networks
2,015
http://arxiv.org/pdf/1508.06904v5
Title Rapid Exact Signal Scanning Deep Convolutional Neural Networks Summary rigorous formulation dynamic signal processing scheme aimed dense signal scanning without loss accuracy introduced analyzed Related method proposed recent past lack satisfactory analysis whether actually fulfill exactness constraint improved e...
[-0.02646723948419094, 0.05041119083762169, -0.02328396588563919, 0.09498894214630127, -0.004298678133636713, -0.019433703273534775, 0.07059383392333984, 0.0016995362238958478, -0.026194630190730095, -0.019928136840462685, -0.008342829532921314, -0.009957071393728256, 0.0157212745398283, 0.03792042285203934, 0.00274867...
1,355
1,355
['Atul Laxman Katole', 'Krishna Prasad Yellapragada', 'Amish Kumar Bedi', 'Sehaj Singh Kalra', 'Mynepalli Siva Chaitanya']
1509.01951v1
Evolution of visual object recognition architectures based on Convolutional Neural Networks & Convolutional Deep Belief Networks paradigms has revolutionized artificial Vision Science. These architectures extract & learn the real world hierarchical visual features utilizing supervised & unsupervised learning approaches...
Hierarchical Deep Learning Architecture For 10K Objects Classification
2,015
http://arxiv.org/pdf/1509.01951v1
Title Hierarchical Deep Learning Architecture 10K Objects Classification Summary Evolution visual object recognition architecture based Convolutional Neural Networks Convolutional Deep Belief Networks paradigm revolutionized artificial Vision Science architecture extract learn real world hierarchical visual feature uti...
[-0.00043182208901271224, 0.05312609300017357, -0.01782836578786373, 0.08956040441989899, 0.003512816969305277, 0.0038716779090464115, 0.03174705058336258, 0.017254451289772987, -0.019054384902119637, -0.013697242364287376, -0.015295793302357197, 0.030945725739002228, -0.030977660790085793, 0.06853853911161423, -0.0183...
1,356
1,356
['Jianwei Luo', 'Jianguo Li', 'Jun Wang', 'Zhiguo Jiang', 'Yurong Chen']
1509.02470v1
Recently, many researches employ middle-layer output of convolutional neural network models (CNN) as features for different visual recognition tasks. Although promising results have been achieved in some empirical studies, such type of representations still suffer from the well-known issue of semantic gap. This paper p...
Deep Attributes from Context-Aware Regional Neural Codes
2,015
http://arxiv.org/pdf/1509.02470v1
Title Deep Attributes ContextAware Regional Neural Codes Summary Recently many research employ middlelayer output convolutional neural network model CNN feature different visual recognition task Although promising result achieved empirical study type representation still suffer wellknown issue semantic gap paper propos...
[-0.022275490686297417, 0.005107360426336527, -0.01624920219182968, 0.07268967479467392, 0.01149737648665905, 0.029419055208563805, 0.05550307780504227, 0.007925288751721382, -0.06528148800134659, -0.04356138035655022, -0.047783542424440384, 0.020130760967731476, 0.021096350625157356, 0.027961265295743942, 0.0127936890...
1,357
1,357
['Wentao Zhu', 'Jun Miao', 'Laiyun Qing', 'Xilin Chen']
1509.08038v1
Learning features from massive unlabelled data is a vast prevalent topic for high-level tasks in many machine learning applications. The recent great improvements on benchmark data sets achieved by increasingly complex unsupervised learning methods and deep learning models with lots of parameters usually requires many ...
Deep Trans-layer Unsupervised Networks for Representation Learning
2,015
http://arxiv.org/pdf/1509.08038v1
Title Deep Translayer Unsupervised Networks Representation Learning Summary Learning feature massive unlabelled data vast prevalent topic highlevel task many machine learning application recent great improvement benchmark data set achieved increasingly complex unsupervised learning method deep learning model lot parame...
[-0.04376518726348877, 0.004774442408233881, -0.010686388239264488, 0.03943752497434616, 0.002097331453114748, 0.003855145303532481, 0.0729469358921051, -0.023324809968471527, -0.03260403871536255, 0.011276094242930412, -0.03226759284734726, 0.028780166059732437, 0.0014909362653270364, 0.07199984043836594, 0.0238978024...
1,358
1,358
['Guillaume Soulié', 'Vincent Gripon', 'Maëlys Robert']
1509.08745v5
Thanks to their state-of-the-art performance, deep neural networks are increasingly used for object recognition. To achieve these results, they use millions of parameters to be trained. However, when targeting embedded applications the size of these models becomes problematic. As a consequence, their usage on smartphon...
Compression of Deep Neural Networks on the Fly
2,015
http://arxiv.org/pdf/1509.08745v5
Title Compression Deep Neural Networks Fly Summary Thanks stateoftheart performance deep neural network increasingly used object recognition achieve result use million parameter trained However targeting embedded application size model becomes problematic consequence usage smartphones resource limited device prohibited...
[-0.037633661180734634, 0.054040130227804184, -0.00485815666615963, 0.026522304862737656, -0.02769269049167633, -0.017377294600009918, 0.06027292460203171, 0.02660202607512474, -0.07784005999565125, 0.02253536507487297, -0.0002531877835281193, 0.03335215151309967, 0.03344383463263512, 0.037558890879154205, 0.0187646187...
1,359
1,359
['Matthieu Courbariaux', 'Yoshua Bengio', 'Jean-Pierre David']
1511.00363v3
Deep Neural Networks (DNN) have achieved state-of-the-art results in a wide range of tasks, with the best results obtained with large training sets and large models. In the past, GPUs enabled these breakthroughs because of their greater computational speed. In the future, faster computation at both training and test ti...
BinaryConnect: Training Deep Neural Networks with binary weights during propagations
2,015
http://arxiv.org/pdf/1511.00363v3
Title BinaryConnect Training Deep Neural Networks binary weight propagation Summary Deep Neural Networks DNN achieved stateoftheart result wide range task best result obtained large training set large model past GPUs enabled breakthrough greater computational speed future faster computation training test time likely cr...
[-0.0191330686211586, 0.022287679836153984, -0.036507345736026764, 0.03610583767294884, -0.014705998823046684, -0.03818388655781746, 0.10037252306938171, -0.01837189868092537, -0.03577876091003418, 0.013395520858466625, 0.03382125869393349, -0.018581025302410126, 0.013095834292471409, 0.07455985993146896, 0.02823296189...
1,360
1,360
['Natalia Neverova', 'Christian Wolf', 'Griffin Lacey', 'Lex Fridman', 'Deepak Chandra', 'Brandon Barbello', 'Graham Taylor']
1511.03908v4
We present a large-scale study exploring the capability of temporal deep neural networks to interpret natural human kinematics and introduce the first method for active biometric authentication with mobile inertial sensors. At Google, we have created a first-of-its-kind dataset of human movements, passively collected b...
Learning Human Identity from Motion Patterns
2,015
http://arxiv.org/pdf/1511.03908v4
Title Learning Human Identity Motion Patterns Summary present largescale study exploring capability temporal deep neural network interpret natural human kinematics introduce first method active biometric authentication mobile inertial sensor Google created firstofitskind dataset human movement passively collected 1500 ...
[-0.015504722483456135, 0.05523623153567314, -0.019333936274051666, 0.034982286393642426, 0.00373988994397223, 0.008848813362419605, 0.04953227564692497, -1.3098569070280064e-05, -0.02989005111157894, -0.037601202726364136, 0.023386558517813683, -0.01972634345293045, 0.016322800889611244, 0.045210279524326324, 0.034482...
1,361
1,361
['Ziming Zhang', 'Yuting Chen', 'Venkatesh Saligrama']
1511.04524v2
In this paper, we propose training very deep neural networks (DNNs) for supervised learning of hash codes. Existing methods in this context train relatively "shallow" networks limited by the issues arising in back propagation (e.e. vanishing gradients) as well as computational efficiency. We propose a novel and efficie...
Efficient Training of Very Deep Neural Networks for Supervised Hashing
2,015
http://arxiv.org/pdf/1511.04524v2
Title Efficient Training Deep Neural Networks Supervised Hashing Summary paper propose training deep neural network DNNs supervised learning hash code Existing method context train relatively shallow network limited issue arising back propagation ee vanishing gradient well computational efficiency propose novel efficie...
[-0.009914598427712917, 0.007793010678142309, 0.0043750605545938015, 0.05547737330198288, -0.016723955050110817, -0.0011554887751117349, 0.09332090616226196, 0.0052092489786446095, 0.010703056119382381, 0.005873889662325382, 0.014347811229526997, 0.006667045410722494, -0.048634618520736694, 0.06632094830274582, -0.0095...
1,362
1,362
['Sara Sabour', 'Yanshuai Cao', 'Fartash Faghri', 'David J. Fleet']
1511.05122v9
We show that the representation of an image in a deep neural network (DNN) can be manipulated to mimic those of other natural images, with only minor, imperceptible perturbations to the original image. Previous methods for generating adversarial images focused on image perturbations designed to produce erroneous class ...
Adversarial Manipulation of Deep Representations
2,015
http://arxiv.org/pdf/1511.05122v9
Title Adversarial Manipulation Deep Representations Summary show representation image deep neural network DNN manipulated mimic natural image minor imperceptible perturbation original image Previous method generating adversarial image focused image perturbation designed produce erroneous class label concentrate interna...
[0.03884347155690193, 0.045718587934970856, -0.022723298519849777, 0.07036052644252777, -0.04117962718009949, -0.021172838285565376, 0.051231708377599716, -0.014405528083443642, -0.031697165220975876, -0.03199499845504761, -0.032277338206768036, 0.05953541770577431, -0.006603263318538666, 0.022288385778665543, 0.075685...
1,363
1,363
['Min Li', 'Sudeep Gaddam', 'Xiaolin Li', 'Yinan Zhao', 'Jingzhe Ma', 'Jian Ge']
1511.05607v2
The pervasive interstellar dust grains provide significant insights to understand the formation and evolution of the stars, planetary systems, and the galaxies, and may harbor the building blocks of life. One of the most effective way to analyze the dust is via their interaction with the light from background sources. ...
Identifying the Absorption Bump with Deep Learning
2,015
http://arxiv.org/pdf/1511.05607v2
Title Identifying Absorption Bump Deep Learning Summary pervasive interstellar dust grain provide significant insight understand formation evolution star planetary system galaxy may harbor building block life One effective way analyze dust via interaction light background source observed extinction curve spectral featu...
[0.040776073932647705, -0.06202775612473488, -0.0008089550537988544, 0.05674951523542404, 0.0054481541737914085, -0.024156661704182625, 0.012445779517292976, 0.0417516827583313, 0.02825135923922062, 0.051565032452344894, 0.014501196332275867, -0.03134099766612053, -0.046551335602998734, 0.06611695140600204, -0.03299642...
1,364
1,364
['Zhibin Liao', 'Gustavo Carneiro']
1511.05635v1
In this paper, we introduce a new deep convolutional neural network (ConvNet) module that promotes competition among a set of multi-scale convolutional filters. This new module is inspired by the inception module, where we replace the original collaborative pooling stage (consisting of a concatenation of the multi-scal...
Competitive Multi-scale Convolution
2,015
http://arxiv.org/pdf/1511.05635v1
Title Competitive Multiscale Convolution Summary paper introduce new deep convolutional neural network ConvNet module promotes competition among set multiscale convolutional filter new module inspired inception module replace original collaborative pooling stage consisting concatenation multiscale filter output competi...
[-0.010207131505012512, 0.030889388173818588, -0.01655527949333191, 0.06199643388390541, -0.0014448255533352494, -0.007983994670212269, 0.07663045823574066, 0.017388885840773582, -0.050629109144210815, -0.007030679378658533, -0.05834263563156128, -0.013446156866848469, -0.0351467989385128, 0.057231638580560684, 0.02068...
1,365
1,365
['Sheng-Yi Bai', 'Sebastian Agethen', 'Ting-Hsuan Chao', 'Winston Hsu']
1511.06104v2
The recent promising achievements of deep learning rely on the large amount of labeled data. Considering the abundance of data on the web, most of them do not have labels at all. Therefore, it is important to improve generalization performance using unlabeled data on supervised tasks with few labeled instances. In this...
Semi-supervised Learning for Convolutional Neural Networks via Online Graph Construction
2,015
http://arxiv.org/pdf/1511.06104v2
Title Semisupervised Learning Convolutional Neural Networks via Online Graph Construction Summary recent promising achievement deep learning rely large amount labeled data Considering abundance data web label Therefore important improve generalization performance using unlabeled data supervised task labeled instance wo...
[-0.0003559791366569698, 0.01703801564872265, 0.0053917900659143925, 0.0495903305709362, 0.005221971310675144, -0.03309701383113861, 0.04603537917137146, -0.021974792703986168, 0.07996843010187149, 0.02322334423661232, -0.013144773431122303, 0.02959662862122059, -0.06526500731706619, 0.10519368201494217, 0.055803500115...
1,366
1,366
['Junghwan Cho', 'Kyewook Lee', 'Ellie Shin', 'Garry Choy', 'Synho Do']
1511.06348v2
The use of Convolutional Neural Networks (CNN) in natural image classification systems has produced very impressive results. Combined with the inherent nature of medical images that make them ideal for deep-learning, further application of such systems to medical image classification holds much promise. However, the us...
How much data is needed to train a medical image deep learning system to achieve necessary high accuracy?
2,015
http://arxiv.org/pdf/1511.06348v2
Title much data needed train medical image deep learning system achieve necessary high accuracy Summary use Convolutional Neural Networks CNN natural image classification system produced impressive result Combined inherent nature medical image make ideal deeplearning application system medical image classification hold...
[0.015247160568833351, 0.04498583823442459, -0.02557196468114853, 0.004405403044074774, 0.010155333206057549, 0.01840953156352043, 0.061760757118463516, 0.054736021906137466, -0.009569007903337479, 0.03272183984518051, -0.014742037281394005, -0.01987915299832821, -0.010037939064204693, 0.04492829740047455, -0.001318782...
1,367
1,367
['Nicolas Ballas', 'Li Yao', 'Chris Pal', 'Aaron Courville']
1511.06432v4
We propose an approach to learn spatio-temporal features in videos from intermediate visual representations we call "percepts" using Gated-Recurrent-Unit Recurrent Networks (GRUs).Our method relies on percepts that are extracted from all level of a deep convolutional network trained on the large ImageNet dataset. While...
Delving Deeper into Convolutional Networks for Learning Video Representations
2,015
http://arxiv.org/pdf/1511.06432v4
Title Delving Deeper Convolutional Networks Learning Video Representations Summary propose approach learn spatiotemporal feature video intermediate visual representation call percept using GatedRecurrentUnit Recurrent Networks GRUsOur method relies percept extracted level deep convolutional network trained large ImageN...
[-0.002140069380402565, 0.01587698794901371, 0.010763132944703102, 0.08439969271421432, 0.005032236222177744, 0.014575610868632793, 0.011538845486938953, 0.0003653374151326716, -0.02454286254942417, -0.08154553174972534, -0.013594390824437141, -0.0149730509147048, -0.009503601118922234, 0.06948446482419968, 0.026081254...
1,368
1,368
["Keiron O'Shea", 'Ryan Nash']
1511.08458v2
The field of machine learning has taken a dramatic twist in recent times, with the rise of the Artificial Neural Network (ANN). These biologically inspired computational models are able to far exceed the performance of previous forms of artificial intelligence in common machine learning tasks. One of the most impressiv...
An Introduction to Convolutional Neural Networks
2,015
http://arxiv.org/pdf/1511.08458v2
Title Introduction Convolutional Neural Networks Summary field machine learning taken dramatic twist recent time rise Artificial Neural Network ANN biologically inspired computational model able far exceed performance previous form artificial intelligence common machine learning task One impressive form ANN architectur...
[0.03236362710595131, 0.023344611749053, -0.025730054825544357, 0.07315732538700104, -0.01015687920153141, -0.011653827503323555, 0.025147799402475357, 0.030056405812501907, -0.05251988768577576, -0.027669137343764305, -0.00160766311455518, 0.0654778778553009, -0.01223669108003378, 0.051733534783124924, 0.0378653593361...
1,369
1,369
['Francisco Massa', 'Bryan Russell', 'Mathieu Aubry']
1512.02497v2
This paper presents an end-to-end convolutional neural network (CNN) for 2D-3D exemplar detection. We demonstrate that the ability to adapt the features of natural images to better align with those of CAD rendered views is critical to the success of our technique. We show that the adaptation can be learned by compositi...
Deep Exemplar 2D-3D Detection by Adapting from Real to Rendered Views
2,015
http://arxiv.org/pdf/1512.02497v2
Title Deep Exemplar 2D3D Detection Adapting Real Rendered Views Summary paper present endtoend convolutional neural network CNN 2D3D exemplar detection demonstrate ability adapt feature natural image better align CAD rendered view critical success technique show adaptation learned compositing rendered view textured obj...
[-0.006889547687023878, 0.03525228798389435, 0.025712409988045692, 0.10000225156545639, 0.021491045132279396, -0.0013151159510016441, -0.018077757209539413, -0.013476106338202953, -0.054761990904808044, -0.010614708997309208, 0.01763126254081726, 0.04589188098907471, -0.01335950382053852, 0.04374580457806587, 0.0353145...
1,370
1,370
['Aaron van den Oord', 'Nal Kalchbrenner', 'Koray Kavukcuoglu']
1601.06759v3
Modeling the distribution of natural images is a landmark problem in unsupervised learning. This task requires an image model that is at once expressive, tractable and scalable. We present a deep neural network that sequentially predicts the pixels in an image along the two spatial dimensions. Our method models the dis...
Pixel Recurrent Neural Networks
2,016
http://arxiv.org/pdf/1601.06759v3
Title Pixel Recurrent Neural Networks Summary Modeling distribution natural image landmark problem unsupervised learning task requires image model expressive tractable scalable present deep neural network sequentially predicts pixel image along two spatial dimension method model discrete probability raw pixel value enc...
[-0.012290353886783123, 0.04953388497233391, -0.017597706988453865, 0.048853885382413864, -0.0180992241948843, -0.02569461241364479, 0.03119402378797531, -0.036565668880939484, -0.06232219189405441, 0.008880890905857086, 0.027696916833519936, -0.006175371818244457, 0.02264973893761635, 0.0457114577293396, 0.04105775803...
1,371
1,371
['Patrick O. Glauner']
1602.00172v2
Inspired by recent successes of deep learning in computer vision, we propose a novel application of deep convolutional neural networks to facial expression recognition, in particular smile recognition. A smile recognition test accuracy of 99.45% is achieved for the Denver Intensity of Spontaneous Facial Action (DISFA) ...
Deep Learning For Smile Recognition
2,016
http://arxiv.org/pdf/1602.00172v2
Title Deep Learning Smile Recognition Summary Inspired recent success deep learning computer vision propose novel application deep convolutional neural network facial expression recognition particular smile recognition smile recognition test accuracy 9945 achieved Denver Intensity Spontaneous Facial Action DISFA databa...
[0.015654679387807846, 0.07491938024759293, -0.0067332130856812, 0.0474054217338562, 0.020976413041353226, 0.022958021610975266, 0.04713821783661842, 0.01739344745874405, 0.059222880750894547, -0.007330560591071844, 0.005680337082594633, 0.030475007370114326, -0.006530069746077061, 0.06774982064962387, 0.04609809815883...
1,372
1,372
['Alexey Dosovitskiy', 'Thomas Brox']
1602.02644v2
Image-generating machine learning models are typically trained with loss functions based on distance in the image space. This often leads to over-smoothed results. We propose a class of loss functions, which we call deep perceptual similarity metrics (DeePSiM), that mitigate this problem. Instead of computing distances...
Generating Images with Perceptual Similarity Metrics based on Deep Networks
2,016
http://arxiv.org/pdf/1602.02644v2
Title Generating Images Perceptual Similarity Metrics based Deep Networks Summary Imagegenerating machine learning model typically trained loss function based distance image space often lead oversmoothed result propose class loss function call deep perceptual similarity metric DeePSiM mitigate problem Instead computing...
[0.0028222249820828438, 0.03779192268848419, -0.015325239859521389, 0.06585425138473511, -0.036108408123254776, 0.0014238395961001515, 0.0588640421628952, 0.012539009563624859, -0.04878494143486023, 0.012437894009053707, -0.007956402376294136, 0.028495121747255325, -0.02457992359995842, 0.06269954144954681, 0.042920183...
1,373
1,373
['Sander Dieleman', 'Jeffrey De Fauw', 'Koray Kavukcuoglu']
1602.02660v2
Many classes of images exhibit rotational symmetry. Convolutional neural networks are sometimes trained using data augmentation to exploit this, but they are still required to learn the rotation equivariance properties from the data. Encoding these properties into the network architecture, as we are already used to doi...
Exploiting Cyclic Symmetry in Convolutional Neural Networks
2,016
http://arxiv.org/pdf/1602.02660v2
Title Exploiting Cyclic Symmetry Convolutional Neural Networks Summary Many class image exhibit rotational symmetry Convolutional neural network sometimes trained using data augmentation exploit still required learn rotation equivariance property data Encoding property network architecture already used translation equi...
[0.017332499846816063, 0.064276784658432, -0.018163368105888367, 0.0589788518846035, -0.009681979194283485, 0.032049015164375305, 0.048179592937231064, -0.014548146165907383, -0.05385105684399605, -0.0038671353831887245, -0.02865976095199585, 0.02032017707824707, 0.010034971870481968, -0.01766204461455345, 0.0093160849...
1,374
1,374
['Babak Saleh', 'Ahmed Elgammal', 'Jacob Feldman']
1602.02865v1
Deep artificial neural networks have made remarkable progress in different tasks in the field of computer vision. However, the empirical analysis of these models and investigation of their failure cases has received attention recently. In this work, we show that deep learning models cannot generalize to atypical images...
The Role of Typicality in Object Classification: Improving The Generalization Capacity of Convolutional Neural Networks
2,016
http://arxiv.org/pdf/1602.02865v1
Title Role Typicality Object Classification Improving Generalization Capacity Convolutional Neural Networks Summary Deep artificial neural network made remarkable progress different task field computer vision However empirical analysis model investigation failure case received attention recently work show deep learning...
[0.023527560755610466, 0.02081672102212906, -0.014501390047371387, 0.0647818073630333, 0.03483668342232704, 0.014560495503246784, 0.03411189466714859, 0.008557574823498726, -0.061501599848270416, 0.016138847917318344, -0.012920284643769264, 0.019986586645245552, 0.0016242058482021093, 0.043056439608335495, 0.0265369098...
1,375
1,375
['Song Wang', 'Dongchun Ren', 'Li Chen', 'Wei Fan', 'Jun Sun', 'Satoshi Naoi']
1602.07373v1
Recently, the deep neural network (derived from the artificial neural network) has attracted many researchers' attention by its outstanding performance. However, since this network requires high-performance GPUs and large storage, it is very hard to use it on individual devices. In order to improve the deep neural netw...
On Study of the Binarized Deep Neural Network for Image Classification
2,016
http://arxiv.org/pdf/1602.07373v1
Title Study Binarized Deep Neural Network Image Classification Summary Recently deep neural network derived artificial neural network attracted many researcher attention outstanding performance However since network requires highperformance GPUs large storage hard use individual device order improve deep neural network...
[-0.0075675975531339645, 0.04148328676819801, -0.02855258248746395, 0.06432520598173141, 0.0034933274146169424, -0.016312479972839355, 0.0538332425057888, 0.016340170055627823, -0.021987827494740486, 0.008744491264224052, 0.002854307647794485, -0.026241635903716087, 0.01134911272674799, 0.04116052761673927, 0.000889977...
1,376
1,376
['Honghao Shan', 'Matthew H. Tong', 'Garrison W. Cottrell']
1602.08486v2
Precortical neural systems encode information collected by the senses, but the driving principles of the encoding used have remained a subject of debate. We present a model of retinal coding that is based on three constraints: information preservation, minimization of the neural wiring, and response equalization. The r...
A Single Model Explains both Visual and Auditory Precortical Coding
2,016
http://arxiv.org/pdf/1602.08486v2
Title Single Model Explains Visual Auditory Precortical Coding Summary Precortical neural system encode information collected sens driving principle encoding used remained subject debate present model retinal coding based three constraint information preservation minimization neural wiring response equalization resulti...
[-0.023485293611884117, -0.0031260584946721792, -0.01215368788689375, -0.019407905638217926, 0.035982370376586914, -0.004754251334816217, -0.03077159821987152, 0.016933506354689598, -0.07354068756103516, 0.022175440564751625, -0.07699859887361526, 0.028558121994137764, 0.03884245827794075, 0.072780080139637, 0.01890014...
1,377
1,377
['Tao Wei', 'Changhu Wang', 'Yong Rui', 'Chang Wen Chen']
1603.01670v2
We present in this paper a systematic study on how to morph a well-trained neural network to a new one so that its network function can be completely preserved. We define this as \emph{network morphism} in this research. After morphing a parent network, the child network is expected to inherit the knowledge from its pa...
Network Morphism
2,016
http://arxiv.org/pdf/1603.01670v2
Title Network Morphism Summary present paper systematic study morph welltrained neural network new one network function completely preserved define emphnetwork morphism research morphing parent network child network expected inherit knowledge parent network also potential continue growing powerful one much shortened tr...
[-0.03260733559727669, 0.03166213259100914, -0.027245299890637398, 0.03551964834332466, 0.013092314824461937, -0.03306017816066742, 0.02879028022289276, -0.021689224988222122, -0.054982274770736694, -0.0007162822294048965, -0.04799489676952362, 0.044300153851509094, -0.01939317397773266, 0.05031365528702736, 0.03723689...
1,378
1,378
['Lucas Beyer', 'Alexander Hermans', 'Bastian Leibe']
1603.02636v2
We introduce the DROW detector, a deep learning based detector for 2D range data. Laser scanners are lighting invariant, provide accurate range data, and typically cover a large field of view, making them interesting sensors for robotics applications. So far, research on detection in laser range data has been dominated...
DROW: Real-Time Deep Learning based Wheelchair Detection in 2D Range Data
2,016
http://arxiv.org/pdf/1603.02636v2
Title DROW RealTime Deep Learning based Wheelchair Detection 2D Range Data Summary introduce DROW detector deep learning based detector 2D range data Laser scanner lighting invariant provide accurate range data typically cover large field view making interesting sensor robotics application far research detection laser ...
[-0.028733113780617714, 0.0184080321341753, -0.01668228767812252, 0.01523068267852068, -0.0066560800187289715, -0.010137461125850677, 0.036407239735126495, -0.0063332379795610905, -0.005052611697465181, -0.0093125831335783, 0.04226204752922058, 0.05251680314540863, 0.011682811193168163, 0.02853783778846264, -0.01183862...
1,379
1,379
['Koen Groenland', 'Sander Bohte']
1603.03657v1
When a Convolutional Neural Network is used for on-the-fly evaluation of continuously updating time-sequences, many redundant convolution operations are performed. We propose the method of Deep Shifting, which remembers previously calculated results of convolution operations in order to minimize the number of calculati...
Efficient forward propagation of time-sequences in convolutional neural networks using Deep Shifting
2,016
http://arxiv.org/pdf/1603.03657v1
Title Efficient forward propagation timesequences convolutional neural network using Deep Shifting Summary Convolutional Neural Network used onthefly evaluation continuously updating timesequences many redundant convolution operation performed propose method Deep Shifting remembers previously calculated result convolut...
[-0.012064773589372635, 0.025094758719205856, 0.009276381693780422, 0.0163564532995224, -0.036767784506082535, -0.04077623039484024, 0.0537453331053257, -0.016819853335618973, -0.11226141452789307, -0.00682729622349143, 0.03721955418586731, -0.03126177564263344, 0.025091931223869324, 0.06216096878051758, -0.02479566633...
1,380
1,380
['Randal S. Olson', 'Jason H. Moore', 'Christoph Adami']
1603.08233v2
Pattern recognition and classification is a central concern for modern information processing systems. In particular, one key challenge to image and video classification has been that the computational cost of image processing scales linearly with the number of pixels in the image or video. Here we present an intellige...
Evolution of active categorical image classification via saccadic eye movement
2,016
http://arxiv.org/pdf/1603.08233v2
Title Evolution active categorical image classification via saccadic eye movement Summary Pattern recognition classification central concern modern information processing system particular one key challenge image video classification computational cost image processing scale linearly number pixel image video present in...
[0.00601628702133894, 0.011522239074110985, -0.034998342394828796, 0.023948149755597115, 0.04782290384173393, 0.012728869915008545, 0.03853093087673187, 0.04788040369749069, -0.026471754536032677, -0.059310443699359894, 0.007833481766283512, 0.010052364319562912, 0.01101045310497284, 0.05616816505789757, 0.016796734184...
1,381
1,381
['Avanti Shrikumar', 'Peyton Greenside', 'Anna Shcherbina', 'Anshul Kundaje']
1605.01713v3
Note: This paper describes an older version of DeepLIFT. See https://arxiv.org/abs/1704.02685 for the newer version. Original abstract follows: The purported "black box" nature of neural networks is a barrier to adoption in applications where interpretability is essential. Here we present DeepLIFT (Learning Important F...
Not Just a Black Box: Learning Important Features Through Propagating Activation Differences
2,016
http://arxiv.org/pdf/1605.01713v3
Title Black Box Learning Important Features Propagating Activation Differences Summary Note paper describes older version DeepLIFT See httpsarxivorgabs170402685 newer version Original abstract follows purported black box nature neural network barrier adoption application interpretability essential present DeepLIFT Lear...
[-0.014967739582061768, 0.03239334002137184, -0.0240299254655838, -0.0005553952651098371, -0.014666994102299213, -0.00791590940207243, 0.03227587789297104, 0.02271128259599209, -0.0356692299246788, 0.011246494017541409, -0.0013658030657097697, 0.037311114370822906, 0.022124454379081726, 0.09507457166910172, 0.012623527...
1,382
1,382
['Chengxi Ye', 'Chen Zhao', 'Yezhou Yang', 'Cornelia Fermuller', 'Yiannis Aloimonos']
1605.02766v3
LightNet is a lightweight, versatile and purely Matlab-based deep learning framework. The idea underlying its design is to provide an easy-to-understand, easy-to-use and efficient computational platform for deep learning research. The implemented framework supports major deep learning architectures such as Multilayer P...
LightNet: A Versatile, Standalone Matlab-based Environment for Deep Learning
2,016
http://arxiv.org/pdf/1605.02766v3
Title LightNet Versatile Standalone Matlabbased Environment Deep Learning Summary LightNet lightweight versatile purely Matlabbased deep learning framework idea underlying design provide easytounderstand easytouse efficient computational platform deep learning research implemented framework support major deep learning ...
[0.01153653021901846, -0.004737409297376871, 0.019147003069519997, 0.029268350452184677, -0.023424867540597916, -0.005779490806162357, 0.05908331647515297, -0.012658629566431046, 0.008770203217864037, 0.00596090592443943, -0.04245015233755112, 0.05886998400092125, -0.006901418324559927, 0.062428079545497894, 0.00800463...
1,383
1,383
['Kin Gwn Lore', 'Daniel Stoecklein', 'Michael Davies', 'Baskar Ganapathysubramanian', 'Soumik Sarkar']
1605.05368v3
Deep learning became the method of choice in recent year for solving a wide variety of predictive analytics tasks. For sequence prediction, recurrent neural networks (RNN) are often the go-to architecture for exploiting sequential information where the output is dependent on previous computation. However, the dependenc...
Deep Action Sequence Learning for Causal Shape Transformation
2,016
http://arxiv.org/pdf/1605.05368v3
Title Deep Action Sequence Learning Causal Shape Transformation Summary Deep learning became method choice recent year solving wide variety predictive analytics task sequence prediction recurrent neural network RNN often goto architecture exploiting sequential information output dependent previous computation However d...
[-0.03693943843245506, 0.023308902978897095, -0.04381338879466057, -0.023272234946489334, -0.020010460168123245, -0.036621011793613434, 0.04118308052420616, -0.02573649398982525, -0.06472320854663849, 0.03641253709793091, 0.03256237879395485, -0.0004443628713488579, -0.0015067927306517959, 0.12205664068460464, 0.030415...
1,384
1,384
['Saurabh Singh', 'Derek Hoiem', 'David Forsyth']
1605.06465v1
We describe Swapout, a new stochastic training method, that outperforms ResNets of identical network structure yielding impressive results on CIFAR-10 and CIFAR-100. Swapout samples from a rich set of architectures including dropout, stochastic depth and residual architectures as special cases. When viewed as a regular...
Swapout: Learning an ensemble of deep architectures
2,016
http://arxiv.org/pdf/1605.06465v1
Title Swapout Learning ensemble deep architecture Summary describe Swapout new stochastic training method outperforms ResNets identical network structure yielding impressive result CIFAR10 CIFAR100 Swapout sample rich set architecture including dropout stochastic depth residual architecture special case viewed regulari...
[-0.015403693541884422, 0.027343135327100754, -0.005950662307441235, 0.04793112352490425, 0.001934945466928184, -0.010078624822199345, 0.028357749804854393, -0.055229656398296356, -0.05482643097639084, 0.01739889569580555, -0.02573470026254654, 0.017288237810134888, 0.01154134888201952, 0.08024388551712036, 0.004951331...
1,385
1,385
['Yani Ioannou', 'Duncan Robertson', 'Roberto Cipolla', 'Antonio Criminisi']
1605.06489v3
We propose a new method for creating computationally efficient and compact convolutional neural networks (CNNs) using a novel sparse connection structure that resembles a tree root. This allows a significant reduction in computational cost and number of parameters compared to state-of-the-art deep CNNs, without comprom...
Deep Roots: Improving CNN Efficiency with Hierarchical Filter Groups
2,016
http://arxiv.org/pdf/1605.06489v3
Title Deep Roots Improving CNN Efficiency Hierarchical Filter Groups Summary propose new method creating computationally efficient compact convolutional neural network CNNs using novel sparse connection structure resembles tree root allows significant reduction computational cost number parameter compared stateoftheart...
[-0.02452683076262474, 0.028481949120759964, -0.0021106188651174307, 0.07228972762823105, 0.006394647061824799, -0.0464293509721756, -0.014101470820605755, 0.03347824886441231, -0.06843872368335724, 0.011210650205612183, -0.025762232020497322, -0.009655728004872799, 0.031359098851680756, 0.005795541685074568, -0.016660...
1,386
1,386
['Sergey Zagoruyko', 'Nikos Komodakis']
1605.07146v4
Deep residual networks were shown to be able to scale up to thousands of layers and still have improving performance. However, each fraction of a percent of improved accuracy costs nearly doubling the number of layers, and so training very deep residual networks has a problem of diminishing feature reuse, which makes t...
Wide Residual Networks
2,016
http://arxiv.org/pdf/1605.07146v4
Title Wide Residual Networks Summary Deep residual network shown able scale thousand layer still improving performance However fraction percent improved accuracy cost nearly doubling number layer training deep residual network problem diminishing feature reuse make network slow train tackle problem paper conduct detail...
[-0.014714086428284645, 0.027770893648266792, -0.004055788740515709, 0.0816788375377655, 0.04036103934049606, -0.011086924932897091, 0.02687753550708294, 0.002382635371759534, -0.01625009998679161, 0.021212009713053703, -0.019702840596437454, 0.02788665145635605, -0.030576396733522415, 0.007905189879238605, -0.00210604...
1,387
1,387
['Osbert Bastani', 'Yani Ioannou', 'Leonidas Lampropoulos', 'Dimitrios Vytiniotis', 'Aditya Nori', 'Antonio Criminisi']
1605.07262v2
Despite having high accuracy, neural nets have been shown to be susceptible to adversarial examples, where a small perturbation to an input can cause it to become mislabeled. We propose metrics for measuring the robustness of a neural net and devise a novel algorithm for approximating these metrics based on an encoding...
Measuring Neural Net Robustness with Constraints
2,016
http://arxiv.org/pdf/1605.07262v2
Title Measuring Neural Net Robustness Constraints Summary Despite high accuracy neural net shown susceptible adversarial example small perturbation input cause become mislabeled propose metric measuring robustness neural net devise novel algorithm approximating metric based encoding robustness linear program show metri...
[-0.027623847126960754, 0.05387906730175018, -0.015181492082774639, 0.006513364613056183, 0.025442365556955338, -0.03236067667603493, 0.015905197709798813, 0.005242270417511463, -0.005849436856806278, 0.012930899858474731, 0.03199639543890953, -0.015172642655670643, 0.029927892610430717, 0.03644094243645668, 0.05452759...
1,388
1,388
['Ludovic Trottier', 'Philippe Giguère', 'Brahim Chaib-draa']
1605.09332v4
Object recognition is an important task for improving the ability of visual systems to perform complex scene understanding. Recently, the Exponential Linear Unit (ELU) has been proposed as a key component for managing bias shift in Convolutional Neural Networks (CNNs), but defines a parameter that must be set by hand. ...
Parametric Exponential Linear Unit for Deep Convolutional Neural Networks
2,016
http://arxiv.org/pdf/1605.09332v4
Title Parametric Exponential Linear Unit Deep Convolutional Neural Networks Summary Object recognition important task improving ability visual system perform complex scene understanding Recently Exponential Linear Unit ELU proposed key component managing bias shift Convolutional Neural Networks CNNs defines parameter m...
[-0.012830820865929127, 0.022934501990675926, 0.0011080644326284528, 0.07815849781036377, 0.033771615475416183, 0.026768920943140984, 0.06934427469968796, -0.0021529062651097775, -0.05563076585531235, 0.009668804705142975, -0.0007429049583151937, -0.004100256599485874, -0.018747711554169655, 0.05512850731611252, 0.0330...
1,389
1,389
['Yoonchang Han', 'Jaehun Kim', 'Kyogu Lee']
1605.09507v3
Identifying musical instruments in polyphonic music recordings is a challenging but important problem in the field of music information retrieval. It enables music search by instrument, helps recognize musical genres, or can make music transcription easier and more accurate. In this paper, we present a convolutional ne...
Deep convolutional neural networks for predominant instrument recognition in polyphonic music
2,016
http://arxiv.org/pdf/1605.09507v3
Title Deep convolutional neural network predominant instrument recognition polyphonic music Summary Identifying musical instrument polyphonic music recording challenging important problem field music information retrieval enables music search instrument help recognize musical genre make music transcription easier accur...
[0.01390052866190672, 0.020433716475963593, -0.005847972817718983, 0.07407350838184357, 0.035155098885297775, -0.03928114473819733, 0.08882278949022293, -0.022265193983912468, -0.04123317077755928, 0.03950470685958862, -0.056952979415655136, 0.03997141122817993, 0.015469728037714958, 0.02813017927110195, 0.038540307432...
1,390
1,390
['Boris Knyazev', 'Erhardt Barth', 'Thomas Martinetz']
1606.00611v2
In visual recognition tasks, such as image classification, unsupervised learning exploits cheap unlabeled data and can help to solve these tasks more efficiently. We show that the recursive autoconvolution operator, adopted from physics, boosts existing unsupervised methods by learning more discriminative filters. We t...
Recursive Autoconvolution for Unsupervised Learning of Convolutional Neural Networks
2,016
http://arxiv.org/pdf/1606.00611v2
Title Recursive Autoconvolution Unsupervised Learning Convolutional Neural Networks Summary visual recognition task image classification unsupervised learning exploit cheap unlabeled data help solve task efficiently show recursive autoconvolution operator adopted physic boost existing unsupervised method learning discr...
[-0.015421235002577305, 0.030324136838316917, 0.002316541038453579, 0.10182454437017441, 0.012798304669559002, 0.010137788020074368, 0.03407498449087143, -0.03672230616211891, 0.019462209194898605, -0.004186635371297598, 0.002579163759946823, 0.04451262578368187, -0.008690348826348782, 0.04557041451334953, 0.0080915372...
1,391
1,391
['Jean-Charles Vialatte', 'Vincent Gripon', 'Grégoire Mercier']
1606.01166v4
Convolutional Neural Networks (CNNs) have become the state-of-the-art in supervised learning vision tasks. Their convolutional filters are of paramount importance for they allow to learn patterns while disregarding their locations in input images. When facing highly irregular domains, generalized convolutional operator...
Generalizing the Convolution Operator to extend CNNs to Irregular Domains
2,016
http://arxiv.org/pdf/1606.01166v4
Title Generalizing Convolution Operator extend CNNs Irregular Domains Summary Convolutional Neural Networks CNNs become stateoftheart supervised learning vision task convolutional filter paramount importance allow learn pattern disregarding location input image facing highly irregular domain generalized convolutional o...
[0.05403973534703255, 0.03268153965473175, -0.0001967168936971575, 0.024216048419475555, -0.012604337185621262, -0.029166603460907936, 0.05282023176550865, 0.010737384669482708, 0.013027996756136417, 0.017574908211827278, -0.009875945746898651, -0.0050835576839745045, -0.030606592074036598, 0.04758421331644058, 0.03757...
1,392
1,392
['Paul Merolla', 'Rathinakumar Appuswamy', 'John Arthur', 'Steve K. Esser', 'Dharmendra Modha']
1606.01981v1
Recent results show that deep neural networks achieve excellent performance even when, during training, weights are quantized and projected to a binary representation. Here, we show that this is just the tip of the iceberg: these same networks, during testing, also exhibit a remarkable robustness to distortions beyond ...
Deep neural networks are robust to weight binarization and other non-linear distortions
2,016
http://arxiv.org/pdf/1606.01981v1
Title Deep neural network robust weight binarization nonlinear distortion Summary Recent result show deep neural network achieve excellent performance even training weight quantized projected binary representation show tip iceberg network testing also exhibit remarkable robustness distortion beyond quantization includi...
[-0.0187813863158226, 0.0539134256541729, -0.016019191592931747, 0.038408610969781876, 0.007665860932320356, -0.0391228087246418, 0.030506843701004982, 0.028892796486616135, -0.0783856213092804, 0.057649191468954086, 0.039996907114982605, -0.008211196400225163, 0.0004250482306815684, 0.04648032411932945, 0.059446897357...
1,393
1,393
['Dmytro Mishkin', 'Nikolay Sergievskiy', 'Jiri Matas']
1606.02228v2
The paper systematically studies the impact of a range of recent advances in CNN architectures and learning methods on the object categorization (ILSVRC) problem. The evalution tests the influence of the following choices of the architecture: non-linearity (ReLU, ELU, maxout, compatibility with batch normalization), po...
Systematic evaluation of CNN advances on the ImageNet
2,016
http://arxiv.org/pdf/1606.02228v2
Title Systematic evaluation CNN advance ImageNet Summary paper systematically study impact range recent advance CNN architecture learning method object categorization ILSVRC problem evalution test influence following choice architecture nonlinearity ReLU ELU maxout compatibility batch normalization pooling variant stoc...
[0.0048497519455850124, 0.025579843670129776, -0.008124032989144325, 0.06044792756438255, 0.016556646674871445, -0.0317179374396801, 0.06333209574222565, 0.008801227435469627, -0.0589560940861702, -0.0011544612934812903, 0.0045453403145074844, 0.03251326456665993, -0.015507478266954422, 0.042727161198854446, 0.03948436...
1,394
1,394
['Shreyas Saxena', 'Jakob Verbeek']
1606.02492v4
Despite the success of CNNs, selecting the optimal architecture for a given task remains an open problem. Instead of aiming to select a single optimal architecture, we propose a "fabric" that embeds an exponentially large number of architectures. The fabric consists of a 3D trellis that connects response maps at differ...
Convolutional Neural Fabrics
2,016
http://arxiv.org/pdf/1606.02492v4
Title Convolutional Neural Fabrics Summary Despite success CNNs selecting optimal architecture given task remains open problem Instead aiming select single optimal architecture propose fabric embeds exponentially large number architecture fabric consists 3D trellis connects response map different layer scale channel sp...
[-0.003114857943728566, 0.02742823027074337, -0.014369962736964226, 0.09354546666145325, 0.006915135309100151, -0.05828556790947914, 0.03705000877380371, -0.03879992663860321, -0.0796101912856102, 0.025443481281399727, 0.003572674235329032, 0.04641973227262497, 0.0039123548194766045, 0.06076894700527191, 0.063221409916...
1,395
1,395
['Chrisantha Fernando', 'Dylan Banarse', 'Malcolm Reynolds', 'Frederic Besse', 'David Pfau', 'Max Jaderberg', 'Marc Lanctot', 'Daan Wierstra']
1606.02580v1
In this work we introduce a differentiable version of the Compositional Pattern Producing Network, called the DPPN. Unlike a standard CPPN, the topology of a DPPN is evolved but the weights are learned. A Lamarckian algorithm, that combines evolution and learning, produces DPPNs to reconstruct an image. Our main result...
Convolution by Evolution: Differentiable Pattern Producing Networks
2,016
http://arxiv.org/pdf/1606.02580v1
Title Convolution Evolution Differentiable Pattern Producing Networks Summary work introduce differentiable version Compositional Pattern Producing Network called DPPN Unlike standard CPPN topology DPPN evolved weight learned Lamarckian algorithm combine evolution learning produce DPPNs reconstruct image main result DP...
[-0.00800623930990696, 0.12442009150981903, -0.028783485293388367, 0.07464304566383362, -0.0047689517959952354, -0.02463567815721035, 0.013507766649127007, 0.006263635121285915, -0.1060599684715271, 0.07460159808397293, 0.013940079137682915, 0.052749618887901306, -0.015679363161325455, 0.09677726030349731, 0.0414533987...
1,396
1,396
['Andrey Zhmoginov', 'Mark Sandler']
1606.04189v2
Deep neural networks have dramatically advanced the state of the art for many areas of machine learning. Recently they have been shown to have a remarkable ability to generate highly complex visual artifacts such as images and text rather than simply recognize them. In this work we use neural networks to effectively ...
Inverting face embeddings with convolutional neural networks
2,016
http://arxiv.org/pdf/1606.04189v2
Title Inverting face embeddings convolutional neural network Summary Deep neural network dramatically advanced state art many area machine learning Recently shown remarkable ability generate highly complex visual artifact image text rather simply recognize work use neural network effectively invert lowdimensional face ...
[-0.017484348267316818, 0.027124928310513496, -0.012019258923828602, 0.07799743115901947, -0.013198708184063435, -0.004320301581174135, 0.01100866962224245, 0.013867844827473164, -0.021687790751457214, 0.05687787011265755, 0.04267800971865654, 0.0369231253862381, 0.009001434780657291, 0.06841085851192474, 0.06958387792...
1,397
1,397
['Ahmed Mamdouh A. Hassanien']
1607.06125v1
In this work we present a state-of-the-art approach for unconstrained natural scene text recognition. We propose a cascade approach that incorporates a convolutional neural network (CNN) architecture followed by a long short term memory model (LSTM). The CNN learns visual features for the characters and uses them with ...
Sequence to sequence learning for unconstrained scene text recognition
2,016
http://arxiv.org/pdf/1607.06125v1
Title Sequence sequence learning unconstrained scene text recognition Summary work present stateoftheart approach unconstrained natural scene text recognition propose cascade approach incorporates convolutional neural network CNN architecture followed long short term memory model LSTM CNN learns visual feature characte...
[0.020487666130065918, 0.04919007420539856, 0.02711860090494156, 0.11231061071157455, -0.035250768065452576, 0.010679557919502258, 0.02113761380314827, 0.05547526478767395, 0.01051875576376915, -0.04257830232381821, 0.010259457863867283, -0.020291628316044807, 0.01272597536444664, 0.0654185563325882, -0.025548761710524...
1,398
1,398
['Itir Onal Ertugrul', 'Mete Ozay', 'Fatos Tunay Yarman Vural']
1607.07695v2
We propose a new framework, called Hierarchical Multi-resolution Mesh Networks (HMMNs), which establishes a set of brain networks at multiple time resolutions of fMRI signal to represent the underlying cognitive process. The suggested framework, first, decomposes the fMRI signal into various frequency subbands using wa...
Hierarchical Multi-resolution Mesh Networks for Brain Decoding
2,016
http://arxiv.org/pdf/1607.07695v2
Title Hierarchical Multiresolution Mesh Networks Brain Decoding Summary propose new framework called Hierarchical Multiresolution Mesh Networks HMMNs establishes set brain network multiple time resolution fMRI signal represent underlying cognitive process suggested framework first decomposes fMRI signal various frequen...
[-0.048039570450782776, 0.05027308315038681, -0.03500717133283615, -0.025519756600260735, 0.021027181297540665, 0.03935092315077782, 0.03272181376814842, 0.01647804118692875, 0.042165424674749374, 0.061469390988349915, -0.04345134645700455, 0.03961687907576561, 0.07028534263372421, 0.043894071131944656, 0.0410220995545...
1,399
1,399
['Dan Hendrycks', 'Kevin Gimpel']
1608.00530v2
Many machine learning classifiers are vulnerable to adversarial perturbations. An adversarial perturbation modifies an input to change a classifier's prediction without causing the input to seem substantially different to human perception. We deploy three methods to detect adversarial images. Adversaries trying to bypa...
Early Methods for Detecting Adversarial Images
2,016
http://arxiv.org/pdf/1608.00530v2
Title Early Methods Detecting Adversarial Images Summary Many machine learning classifier vulnerable adversarial perturbation adversarial perturbation modifies input change classifier prediction without causing input seem substantially different human perception deploy three method detect adversarial image Adversaries ...
[0.05352093279361725, 0.021090984344482422, -0.03731386363506317, 0.02465878054499626, -0.0010661063715815544, -0.005470412317663431, 0.026511842384934425, 0.005893063265830278, -0.0005782980588264763, -0.04843335598707199, 0.053108058869838715, 0.03659062087535858, 0.008139259181916714, 0.009170649573206902, 0.0407419...