Timezone: »
Data-Augmentation (DA) is known to improve performance across tasks and datasets. We propose a method to theoretically analyze the effect of DA and study questions such as: how many augmented samples are needed to correctly estimate the information encoded by that DA? How does the augmentation policy impact the final parameters of a model? We derive several quantities in close-form, such as the expectation and variance of an image, loss, and model's output under a given DA distribution. Up to our knowledge, we obtain the first explicit regularizer that corresponds to using DA during training for non-trivial transformations such as affine transformations, color jittering, or Gaussian blur. Those derivations open new avenues to quantify the benefits and limitations of DA. For example, given a loss at hand, we find that common DAs require tens of thousands of samples for the loss to be correctly estimated and for the model training to converge. We then show that for a training loss to have reduced variance under DA sampling, the model's saliency map (gradient of the loss with respect to the model's input) must align with the smallest eigenvector of the sample's covariance matrix under the considered DA augmentation; this is exactly the quantity estimated and regularized by TangentProp. Those findings also hint at a possible explanation on why models tend to shift their focus from edges to textures when specific DAs are employed.
Author Information
Randall Balestriero (Meta AI, FAIR)
Ishan Misra (Facebook AI Research)
Yann LeCun (Facebook)
Yann LeCun is Director of AI Research at Facebook, and Silver Professor of Data Science, Computer Science, Neural Science, and Electrical Engineering at New York University. He received the Electrical Engineer Diploma from ESIEE, Paris in 1983, and a PhD in Computer Science from Université Pierre et Marie Curie (Paris) in 1987. After a postdoc at the University of Toronto, he joined AT&T Bell Laboratories in Holmdel, NJ in 1988. He became head of the Image Processing Research Department at AT&T Labs-Research in 1996, and joined NYU as a professor in 2003, after a brief period as a Fellow of the NEC Research Institute in Princeton. From 2012 to 2014 he directed NYU's initiative in data science and became the founding director of the NYU Center for Data Science. He was named Director of AI Research at Facebook in late 2013 and retains a part-time position on the NYU faculty. His current interests include AI, machine learning, computer perception, mobile robotics, and computational neuroscience. He has published over 180 technical papers and book chapters on these topics as well as on neural networks, handwriting recognition, image processing and compression, and on dedicated circuits for computer perception.
More from the Same Authors
-
2022 : Exact Visualization of Deep Neural Network Geometry and Decision Boundary »
Ahmed Imtiaz Humayun · Randall Balestriero · Richard Baraniuk -
2023 Workshop: 4th Workshop on Self-Supervised Learning: Theory and Practice »
Tengda Han · Ishan Misra · Pengtao Xie · Mathilde Caron · Hilde Kuehne -
2022 Workshop: Self-Supervised Learning: Theory and Practice »
Ishan Misra · Pengtao Xie · Gul Varol · Yale Song · Yuki Asano · Xiaolong Wang · Pauline Luc -
2022 Poster: The Effects of Regularization and Data Augmentation are Class Dependent »
Randall Balestriero · Leon Bottou · Yann LeCun -
2022 Poster: VICRegL: Self-Supervised Learning of Local Visual Features »
Adrien Bardes · Jean Ponce · Yann LeCun -
2022 Poster: Coarse-to-Fine Vision-Language Pre-training with Fusion in the Backbone »
Zi-Yi Dou · Aishwarya Kamath · Zhe Gan · Pengchuan Zhang · Jianfeng Wang · Linjie Li · Zicheng Liu · Ce Liu · Yann LeCun · Nanyun Peng · Jianfeng Gao · Lijuan Wang -
2022 Poster: Pre-Train Your Loss: Easy Bayesian Transfer Learning with Informative Priors »
Ravid Shwartz-Ziv · Micah Goldblum · Hossein Souri · Sanyam Kapoor · Chen Zhu · Yann LeCun · Andrew Wilson -
2022 Poster: projUNN: efficient method for training deep networks with unitary matrices »
Bobak Kiani · Randall Balestriero · Yann LeCun · Seth Lloyd -
2022 Poster: Contrastive and Non-Contrastive Self-Supervised Learning Recover Global and Local Spectral Embedding Methods »
Randall Balestriero · Yann LeCun -
2021 Workshop: 2nd Workshop on Self-Supervised Learning: Theory and Practice »
Pengtao Xie · Ishan Misra · Pulkit Agrawal · Abdelrahman Mohamed · Shentong Mo · Youwei Liang · Jeannette Bohg · Kristina N Toutanova -
2021 Poster: Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers »
Mandela Patrick · Dylan Campbell · Yuki Asano · Ishan Misra · Florian Metze · Christoph Feichtenhofer · Andrea Vedaldi · João Henriques -
2021 Oral: Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers »
Mandela Patrick · Dylan Campbell · Yuki Asano · Ishan Misra · Florian Metze · Christoph Feichtenhofer · Andrea Vedaldi · João Henriques -
2020 : Panel Discussion & Closing »
Yejin Choi · Alexei Efros · Chelsea Finn · Kristen Grauman · Quoc V Le · Yann LeCun · Ruslan Salakhutdinov · Eric Xing -
2020 : QA: Yann LeCun »
Yann LeCun -
2020 : Invited Talk: Yann LeCun »
Yann LeCun -
2020 Workshop: Self-Supervised Learning -- Theory and Practice »
Pengtao Xie · Shanghang Zhang · Pulkit Agrawal · Ishan Misra · Cynthia Rudin · Abdelrahman Mohamed · Wenzhen Yuan · Barret Zoph · Laurens van der Maaten · Xingyi Yang · Eric Xing -
2020 Poster: Analytical Probability Distributions and Exact Expectation-Maximization for Deep Generative Networks »
Randall Balestriero · Sebastien PARIS · Richard Baraniuk -
2019 : TBD »
Yann LeCun -
2019 Poster: The Geometry of Deep Networks: Power Diagram Subdivision »
Randall Balestriero · Romain Cosentino · Behnaam Aazhang · Richard Baraniuk -
2017 : Panel Session »
Neil Lawrence · Finale Doshi-Velez · Zoubin Ghahramani · Yann LeCun · Max Welling · Yee Whye Teh · Ole Winther -
2017 Tutorial: Geometric Deep Learning on Graphs and Manifolds »
Michael Bronstein · Joan Bruna · arthur szlam · Xavier Bresson · Yann LeCun -
2016 : Discussion panel »
Ian Goodfellow · Soumith Chintala · Arthur Gretton · Sebastian Nowozin · Aaron Courville · Yann LeCun · Emily Denton -
2016 : Energy-Based Adversarial Training and Video Prediction »
Yann LeCun -
2016 Symposium: Deep Learning Symposium »
Yoshua Bengio · Yann LeCun · Navdeep Jaitly · Roger Grosse -
2015 Poster: Learning to Linearize Under Uncertainty »
Ross Goroshin · Michael Mathieu · Yann LeCun -
2015 Poster: Character-level Convolutional Networks for Text Classification »
Xiang Zhang · Junbo (Jake) Zhao · Yann LeCun -
2015 Poster: Deep learning with Elastic Averaging SGD »
Sixin Zhang · Anna Choromanska · Yann LeCun -
2015 Spotlight: Deep learning with Elastic Averaging SGD »
Sixin Zhang · Anna Choromanska · Yann LeCun -
2015 Tutorial: Deep Learning »
Geoffrey E Hinton · Yoshua Bengio · Yann LeCun