Relative stability toward diffeomorphisms indicates performance in deep nets

Abstract

Understanding why deep nets can classify data in large dimensions remains a challenge. It has been proposed that they do so by becoming stable to diffeomorphisms, yet existing empirical measurements support that it is often not the case. We revisit this question by defining a maximum-entropy distribution on diffeomorphisms, that allows to study typical diffeomorphisms of a given norm. We confirm that stability toward diffeomorphisms does not strongly correlate to performance on benchmark data sets of images. By contrast, we find that the stability toward diffeomorphisms relative to that of generic transformations $R_f$ correlates remarkably with the test error ${\epsilon }_t$. It is of order unity at initialization but decreases by several decades during training for state-of-the-art architectures. For CIFAR10 and 15 known architectures, we find ${\epsilon }_t\approx 0.2\sqrt{R_f}$, suggesting that obtaining a small $R_f$ is important to achieve good performance. We study how $R_f$ depends on the size of the training set and compare it to a simple model of invariant learning.

arXiv commentNeurIPS 2021 Conference

References in corpus (33)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Playing Atari with Deep Reinforcement Learning
• Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
• SGDR: Stochastic Gradient Descent with Warm Restarts
• Reconciling modern machine learning practice and the bias-variance trade-off
• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Opening the Black Box of Deep Neural Networks via Information
• Understanding Deep Convolutional Networks
• Synthesizing Robust Adversarial Examples
• An Empirical Evaluation of Deep Learning on Highway Driving
• Making Convolutional Networks Shift-Invariant Again
• Breaking the Curse of Dimensionality with Convex Neural Networks
• Sensitivity and Generalization in Neural Networks: an Empirical Study
• Exploring the Landscape of Spatial Robustness
• On Lazy Training in Differentiable Programming
• Spatially Transformed Adversarial Examples
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Intrinsic dimension of data representations in deep neural networks
• A Fourier Perspective on Model Robustness in Computer Vision
• When Do Neural Networks Outperform Kernel Methods?
• Disentangling feature and lazy training in deep neural networks
• On the Inductive Bias of Neural Tangent Kernels
• Dimensionality compression and expansion in Deep Neural Networks
• Finite Versus Infinite Neural Networks: an Empirical Study
• Generalization Guarantees for Neural Networks via Harnessing the Low-rank Structure of the Jacobian
• Implicit Bias of Gradient Descent for Wide Two-layer Neural Networks Trained with the Logistic Loss
• Classifying high-dimensional Gaussian mixtures: Where kernel methods fail and neural networks succeed
• ADef: an Iterative Algorithm to Construct Adversarial Deformations
• Geometric compression of invariant manifolds in neural nets
• Pooling is neither necessary nor sufficient for appropriate deformation stability in CNNs
• Perspective: A Phase Diagram for Deep Learning unifying Jamming, Feature Learning and Lazy Training