Yann LeCun's Publications

MENU

Papers on ArXiv.org: latest results, some of which may or may not make it to a conference or journal. This may contain some of our most recent and interesting results.
Publication Highlights: selected papers on diverse topics with short descriptions.
Full List of Publications: all published papers in reverse chronological order. Downloadable in PDF, DjVu, and gzipped PostScript.
Full List of Papers from Google Scholar: This contains all of my papers that Google Scholar has indexed. They can be displayed in reverse chronological order or in decreasing order of number of citations.

BibTeX file of all papers: so you can easily cite our papers ;-)
BibTeX file of all tech reports: so you can easily cite them.
Full Text Search through PDF papers: search through all PDF papers.
Full Text Search through DjVu papers: search through all DjVu papers.
Download DjVu Viewer: DjVu files are smaller and display much faster than PDF. If you are using Linux, your default document viewer already supports DjVu.

Selected List of Representative Papers

[Farabet et al. 2013]: Learning Hierarchical Features for Scene Labeling, scheduled to appear in the special issue on deep learning of IEEE Trans. on Pattern Analysis and Machine Intelligence. The task is to label all the pixels in an image with the category of the object it belongs to. This is sometimes called scene labeling, scene parsing, or semantic segmentation. The bottom line is that our system beat all previously published scene labeling systems on accuracy on three standard datasets: Stanford Bakground (8 classes), SIFTflow (33 classes) and Barcelona (170 classes). It alsod beat the best competitors by a factor of 100 in speed. Our system is a multiscale convolutional network trained in purely supervised mode (with backprop) to label each pixel. The decisions are then cleaned up by a simple post-processing (the simplest one consisting in taking the majority category within a superpixel).

[Hadsell et al. 2009]: Learning Long-Range Vision for Autonomous Off-Road Driving, and a companion paper [Sermanet et al. 2009]:A Multi-Range Architecture for Collision-Free Off-Road Robot Navigation both scheduled to appear in the Journal of Field Robotics: These two papers describe (in excruciating details) our work on the DARPA LAGR project. We developed a learning-based long-range vision system that can detect obstacles and pathways at very long range, using a combination of training from log files in the lab and on-line adaptation as the robot runs. The robot uses labels obtained from stereo vision to train its monocular long-range obstacle classifier. The system also uses learning for it dynamical trajectory control. Further information is available here.

[LeCun et al. 2006]: A Tutorial on Energy-Based Learning (in Bakir et al. (eds) "Predicting Strutured Data", MIT Press 2006): This is a tutorial paper on Energy-Based Models (EBM). Inference in EBMs consists in searching for the value of the output variables that minimize an energy function. Learning consists in shaping that energy function in such a way that desired configuration have lower energy than undesired ones. The EBM approach provides a common theoretical framework for many probabilistic and non-probabilistic learning models, including traditional discriminative and generative approaches, as well as graph-transformer networks, conditional random fields, maximum margin Markov networks, and several manifold learning methods. Some of the methods described in this paper help circumvent the problem of evaluating partition functions that often plagues probabilistic methods. Further information is available here.

[Bengio, LeCun 2007]: Scaling Learning Algorithms Towards AI: (in Bottou et al. (Eds) "Large-Scale Kernel Machines", MIT Press 2007). We present theoretical and empirical evidence showing that kernel methods and other "shallow" architectures are inefficient for representing complex functions such as the ones involved in artificially intelligent behavior, such as visual perception. We argue that "deep" architectures are not subject to the same limitations and review recent advances in learning algorithms for deep architectures.

[Mirowski et al., 2008]: Comparing SVM and Convolutional Networks for Epileptic Seizure Prediction from Intracranial EEG (MLSP 2008): We show that epilepsy seizures can be predicted about one hour in advance, with essentially no false positives, using signals from intracranial electrodes. A number of different pairwise features that measure the synchrony between pairs of electrodes over 5-second time segments were used. Temporal Convolutional Networks and Support Vector Machines fed with 1-minute sequences of feature vectors were tested the Freiburg dataset. The convolutional network was shown to detect all seizures about 1 hour in advance with no false alarm for all patients in the dataset, significantly outperforming the SVM.

[Chopra et al., 2007]: Discovering the hidden structure of house prices with non-parametric latent manifold model (KDD 2007): In many regression problems, the variable to be predicted depends not only on a sample-specific feature vector, but also on an unknown (latent) manifold that must satisfy known constraints. An example is house prices which depend on the characteristics of the house, and on the desirability of the neighborhood, which is not directly measurable. The proposed method comprises two trainable components. The first one is a parametric model that predicts the "intrinsic" price the house from its description. The second one is a smooth, non-parametric model of the latent "desirability" manifold. The predicted price of a house is the product intrinsic price and desirability. The two components are trained simultanesously using a deterministic form of the EM algorithm. The model was trained on a large dataset of house prices from Los Angeles county. It produces better predictions than pure parametric and non-parametric models. It also produces useful estimates of the desirability surface at each location.

[LeCun, Huang, and Bottou, 2004]: Learning Methods for Generic Object Recognition with Invariance to Pose and Lighting (CVPR 2004): Generic object detection and recognition using convolutional nets. The system can detect and recognize cars, truck, airplanes, human figures, and 4-legged animals in cluttered scenes in real time, with invariance to pose, illumination and clutter. Further information is available here.

[Osadchy, Miller and LeCun, 2007] and [Osadchy, Miller and LeCun, 2005]: Synergistic Face Detection and Pose Estimation with Energy-Based Model (NIPS 2004, JMLR 2007): real-time simultaneous face detection and pose estimation with convolutional networks trained to produce points on a "face manifold" using an Energy-Based loss function. Further information is available here. The method is a direct descendent of the first learning-based system for face detection [Vaillant, Monrocq, and LeCun 1994]: Original approach for the localisation of objects in images, IEE Proc on Vision, Image, and Signal Processing (1994), which followed a paper with the same title published at ICANN 1993 (these predate [Rowley, Baluja, Kanade 1997] and [Viola and Jones 2001]).

[Hadsell, Chopra and LeCun, 2006]: Dimensionality Reduction by Learning an Invariant Mapping (CVPR 2006): Mapping image to a low dimensional representation with invariance to illumination or other factors.

[Chopra and Hadsell and LeCun, 2005]: Learning a Similarity Metric Discriminatively, with Application to Face Verification (CVPR 2005): Using convolutional nets and Energy-Based Models to learn an invariant similarity metric between images of faces.

[LeCun et al., 2005]: Off-Road Obstacle Avoidance through End-to-End Learning (NIPS 2005): Training a robot to avoid obstacles by watching over the shoulder of a human driver. No preprocessing necessary. Further information and videos are available here

[Ning et al., 2005]: Toward Automatic Phenotyping of Developing Embryos from Videos (IEEE Trans. Image Processing, 2005): Using convolutional nets and Energy-Based Models to segment and locate the cells and nuclei in videos of developing embryos of C. Elegans roundworms.

[LeCun et al., 1998]: Gradient-Based Learning Applied to Document Recognition (Proc. IEEE 1998): A long and detailed paper on convolutional nets, graph transformer networks, and discriminative training methods for sequence labeling. We show how to build systems that integrate segmentation, feature extraction, classification, contextual post-processing, and language modeling into one single learning machine trained end-to-end. Applications to handwriting recognition and face detection are described.

[Simard et al., 2000]: Transformation Invariance in Pattern Recognition: Our latest and most complete paper on Tangent Distance, a method for making distance-based classifiers (nearest neighbor, SVM,...) locally invariant to a set of known transformation, and Tangent Propagation, a method for training learning machines to be locally invariant to a set of transformations.

[LeCun et al., 1998]: Efficient BackProp: all the tricks and the theory behind them to efficiently train neural networks with backpropagation, including how to compute the optimal learning rate, how to back-propagate second derivatives, and other sundries.

[Bottou et al., 1998]: High Quality Document Image Compression with DjVu: our first paper on DjVu, still the best method for compressing and distributing scanned documents and high-resolution images. the results are a bit dated, but the more recent papers are less all-encompassing.

[LeCun, 1988]: A theoretical framework for Back-Propagation: how backprop for feed-forward and recurrent nets can be derived cleanly from a Lagrangian formalism.

[LeCun, Denker, and Solla, 1990]: Optimal Brain Damage: a simple and effective "pruning" technique to remove extraneous parameters in learning machines.

[Simard and LeCun, 1992] Reverse TDNN: an architecture for trajectory generation: how a convolutional net turned on its head can be used to synthesize signals or images, instead of recognizing them.

[LeCun and Kanter and Solla, 1991]: Eigenvalues of covariance matrices: application to neural-network learning: why learning is optimal when the number of training samples is about 4 times the number of parameters.

All Publications (in reverse chronological order)

Download BibTeX file

generated by bib2html on 2014-06-21 at 17:38:45

NODES

coding ⁴

Handwritten Digits ¹

Interesting ¹

Intern ⁶⁸

mac ²³

multimedia ²

Note ⁶

os ⁹⁴

text ³

Training ¹⁰

visual ⁵

web ²

187.	Camille Couprie, Clement Farabet, Laurent Najman and Yann LeCun: Toward Real-time Indoor Semantic Segmentation Using Depth Information, JMLR, to appear. Video, 2014, \cite{couprie-jmlr-14}.	1161KB	DjVu
		11222KB	PDF
		19994KB	PS.GZ

186.	Jonatan Tompson, Murphy Stein, Yann LeCun and Ken Perlin: Real-Time Continuous Pose Recovery of Human Hands Using Convolutional Networks , ACM Transaction on Graphics, to appear. Video, 2014, \cite{tompson-siggraph-14}.	837KB	DjVu
		7885KB	PDF
		7962KB	PS.GZ

185.	Joan Bruna, Arthur Szlam and Yann LeCun: Signal Recovery from Lp Pooling Representations, International Conference on Machine Learning (ICML'14), 2014, \cite{bruna-icml-14}.	279KB	DjVu
		269KB	PDF
		289KB	PS.GZ

184.	Pierre Sermanet, David Eigen, Xiang Zhang, Michael Mathieu, Rob Fergus and Yann LeCun: OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks, International Conference on Learning Representations (ICLR2014), CBLS, (OpenReview), (Arxiv:1312.6229); Video of the talk, April 2014, \cite{sermanet-iclr-14}.	451KB	DjVu
		937KB	PDF
		918KB	PS.GZ

183.	Joan Bruna, Wojciech Zaremba, Arthur Szlam and Yann LeCun: Spectral Networks and Locally Connected Networks on Graphs, International Conference on Learning Representations (ICLR2014), CBLS, (OpenReview), (arXiv:1312.6203), April 2014, \cite{bruna-iclr-14}.	421KB	DjVu
		1949KB	PDF
		2586KB	PS.GZ

182.	Michael Mathieu, Mikael Henaff and Yann LeCun: Fast Training of Convolutional Networks through FFTs, International Conference on Learning Representations (ICLR2014), CBLS, (OpenReview), (arXiv:1312.5851), April 2014, \cite{mathieu-iclr-14}.	102KB	DjVu
		338KB	PDF
		327KB	PS.GZ

181.	David Eigen, Jason Rolfe, Rob Fergus and Yann LeCun: Understanding Deep Architectures using a Recursive Convolutional Network, International Conference on Learning Representations (ICLR2014), CBLS, (OpenReview), (arXiv:1312.1847), April 2014, \cite{eigen-iclr-14}.	136KB	DjVu
		556KB	PDF
		316KB	PS.GZ

180.	EJ Humphrey, JP Bello and Y LeCun: Feature learning and deep architectures: new directions for music informatics, Journal of Intelligent Information Systems, 41(3):461-481, 2013, \cite{humphrey-13}.	1190KB	DjVu
		1169KB	PDF
		2814KB	PS.GZ

179.	Camille Couprie, Clement Farabet, Yann LeCun and Laurent Najman: Causal Graph-Based Video Segmentation, Proc. International Conference on Image Processing (ICIP'13), IEEE, September 2013, \cite{couprie-icip-13}.	509KB	DjVu
		2815KB	PDF
		2629KB	PS.GZ

178.	Pierre Sermanet, Koray Kavukcuoglu, Soumith Chintala and Yann LeCun: Pedestrian Detection with Unsupervised Multi-Stage Feature Learning, Proc. International Conference on Computer Vision and Pattern Recognition (CVPR'13), IEEE, Video part 1; Video part 2, June 2013, \cite{sermanet-cvpr-13}.	208KB	DjVu
		396KB	PDF
		476KB	PS.GZ

177.	Tom Schaul, Sixin Zhang and Yann LeCun: No more Pesky Learning Rates, Proc. International Conference on Machine learning (ICML'13), 2013, \cite{schaul-icml-13}.	738KB	DjVu
		1637KB	PDF
		2334KB	PS.GZ

176.	Li Wan, Matthew Zeiler, Sixin Zhang, Yann LeCun and Rob Fergus: Regularization of Neural Networks using DropConnect, Proc. International Conference on Machine learning (ICML'13), 2013, \cite{wan-icml-13}.	396KB	DjVu
		1058KB	PDF
		1058KB	PS.GZ

175.	Jason Tyler Rolfe and Yann LeCun: Discriminative Recurrent Sparse Auto-Encoders, International Conference on Learning Representations (ICLR2013), April 2013, \cite{rolfe-lecun-iclr-13}.	368KB	DjVu
		724KB	PDF
		900KB	PS.GZ

174.	Rotislav Goroshin and Yann LeCun: Saturating Auto-Encoders, International Conference on Learning Representations (ICLR2013), April 2013, \cite{goroshin-lecun-iclr-13}.	229KB	DjVu
		1088KB	PDF
		3918KB	PS.GZ

173.	Camille Couprie, Clement Farabet, Laurent Najman and Yann LeCun: Indoor Semantic Segmentation using Depth Information, International Conference on Learning Representations (ICLR2013), April 2013, \cite{couprie-iclr-13}.	393KB	DjVu
		7578KB	PDF
		12238KB	PS.GZ

172.	Clement Farabet, Camille Couprie, Laurent Najman and Yann LeCun: Learning Hierarchical Features for Scene Labeling, IEEE Transactions on Pattern Analysis and Machine Intelligence, August 2013, \cite{farabet-pami-13}.	1144KB	DjVu
		1432KB	PDF
		1766KB	PS.GZ

171.	Eric J. Humphrey, Juan Pablo Bello and Yann LeCun: Moving beyond feature design: Deep architectures and automatic feature learning in music informatics, Proceedings of International Symposium on Music Information Retrieval (ISMIR'12), 2012, \cite{humphrey-ismir-12}.	368KB	DjVu
		645KB	PDF
		773KB	PS.GZ

170.	Pierre Sermanet, Soumith Chintala and Yann LeCun: Convolutional Neural Networks Applied to House Numbers Digit Classification, International Conference on Pattern Recognition (ICPR 2012), 2012, \cite{sermanet-icpr-12}.	234KB	DjVu
		468KB	PDF
		576KB	PS.GZ

169.	Yann LeCun: Learning Invariant Feature Hierarchies, in Fusiello, Andrea and Murino, Vittorio and Cucchiara, Rita (Eds), European Conference on Computer Vision (ECCV 2012), 7583:496-505, Lecture Notes in Computer Science, Springer, ISBN:978-3-642-33862-5, Workshop on Biological and Computer Vision Interfaces (invited paper), 2012, \cite{lecun-eccv-12}.	284KB	DjVu
		564KB	PDF
		608KB	PS.GZ

168.	Arthur Szlam, Karol Gregor and Yann LeCun: Fast Approximations to Structured Sparse Coding and Applications to Object Classification, in Fitzgibbon, Andrew and Lazebnik, Svetlana and Perona, Pietro and Sato, Yoichi and Schmid, Cordelia (Eds), European Conference on Computer Vision (ECCV 2012), 7576:200-213, Lecture Notes in Computer Science, Springer, ISBN:978-3-642-33714-7, 2012, \cite{szlam-gregor-lecun-eccv-12}.	302KB	DjVu
		276KB	PDF
		281KB	PS.GZ

167.	Jose M. Alvarez, Theo Gevers, Yann LeCun and Antonio M. Lopez: Road Scene Segmentation from a Single Image, in Fitzgibbon, Andrew and Lazebnik, Svetlana and Perona, Pietro and Sato, Yoichi and Schmid, Cordelia (Eds), European Conference on Computer Vision (ECCV 2012), 7578:376-389, Lecture Notes in Computer Science, Springer, ISBN:978-3-642-33785-7, 2012, \cite{alvarez-eccv-12}.	736KB	DjVu
		2722KB	PDF
		3366KB	PS.GZ

166.	Jose M. Alvarez, Yann LeCun, Theo Gevers and Antonio M. Lopez: Semantic Road Segmentation via Multi-scale Ensembles of Learned Features, in Fusiello, Andrea and Murino, Vittorio and Cucchiara, Rita (Eds), European Conference on Computer Vision (ECCV 2012), 7584:586-595, Lecture Notes in Computer Science, Springer, ISBN:978-3-642-33867-0, Workshop on Computer Vision in Vehicle Technology: From Earth to Mars, 2012, \cite{alvarez-eccv-12b}.	595KB	DjVu
		2798KB	PDF
		3957KB	PS.GZ

165.	P. Mirowski and Y. LeCun: Statistical Machine Learning and Dissolved Gas Analysis: A Review, IEEE Transactions on Power Delivery, 27(4):1791-1799, october 2012, \cite{mirowski-lecun-12}.	357KB	DjVu
		780KB	PDF
		682KB	PS.GZ

164.	Phi-Hung Pham, Darko Jelaca, Clement Farabet, Berin Martini, Yann LeCun and Eugenio Culurciello: NeuFlow: Dataflow Vision Processing System-on-a-Chip, Proc. International Midwest Symposium on Circuits and Systems (MWSCAS'12), IEEE, invited paper, 2012, \cite{pham-mwscas-12}.	514KB	DjVu
		1718KB	PDF
		1728KB	PS.GZ

163.	Clement Farabet, Camille Couprie, Laurent Najman and Yann LeCun: Scene Parsing with Multiscale Feature Learning, Purity Trees, and Optimal Covers, Proc. International Conference on Machine learning (ICML'12), 2012, \cite{farabet-icml-12}.	1009KB	DjVu
		2008KB	PDF
		3492KB	PS.GZ

162.	Clement Farabet, Rafael Paz, Jose Perez-Carrasco, Carlos Zamarreno, Alejandro Linares-Barranco, Yann LeCun, Eugenio Culurciello, Teresa Serrano-Gotarredona and Bernabe Linares-Barranco: Comparison Between Frame-Constrained Fix-Pixel-Value and Frame-Free Spiking-Dynamic-Pixel ConvNets for Visual Processing, Frontiers in Neuroscience, 6(00032), DOI: 10.3389/fnins.2012.00032 (open access), 2012, \cite{farabet-frontiersin-12}.	939KB	DjVu
		2073KB	PDF
		1554KB	PS.GZ

161.	Tapani Raiko, Harri Valpola and Yann LeCun: Deep Learning Made Easier by Linear Transformations in Perceptrons, Conference on AI and Statistics (JMLR W&CP), 22:924-932, (JMLR link), 2012, \cite{raiko-aistats-12}.	274KB	DjVu
		881KB	PDF
		1044KB	PS.GZ

160.	Karol Gregor, Arthur Szlam and Yann LeCun: Structured Sparse Coding via Lateral Inhibition, Advances in Neural Information Processing Systems (NIPS 2011), 24, 2011, \cite{gregor-nips-11}.	321KB	DjVu
		240KB	PDF
		252KB	PS.GZ

159.	Mikael Henaff, Kevin Jarrett, Koray Kavukcuoglu and Yann LeCun: Unsupervised Learning of Sparse Features for Scalable Audio Classification, Proceedings of International Symposium on Music Information Retrieval (ISMIR'11), (Best Student Paper Award), 2011, \cite{henaff-ismir-11}.	147KB	DjVu
		297KB	PDF
		311KB	PS.GZ

158.	Y-Lan Boureau, Nicolas Le Roux, Francis Bach, Jean Ponce and Yann LeCun: Ask the locals: multi-way local pooling for image recognition, Proc. International Conference on Computer Vision (ICCV'11), 2011, \cite{boureau-iccv-11}.	130KB	DjVu
		300KB	PDF
		255KB	PS.GZ