Uniform Convergence, Adversarial Spheres and a Simple Remedy

Abstract

Previous work has cast doubt on the general framework of uniform convergence and its ability to explain generalization in neural networks. By considering a specific dataset, it was observed that a neural network completely misclassifies a projection of the training data (adversarial set), rendering any existing generalization bound based on uniform convergence vacuous. We provide an extensive theoretical investigation of the previously studied data setting through the lens of infinitely-wide models. We prove that the Neural Tangent Kernel (NTK) also suffers from the same phenomenon and we uncover its origin. We highlight the important role of the output bias and show theoretically as well as empirically how a sensible choice completely mitigates the problem. We identify sharp phase transitions in the accuracy on the adversarial set and study its dependency on the training sample size. As a result, we are able to characterize critical sample sizes beyond which the effect disappears. Moreover, we study decompositions of a neural network into a clean and noisy part by considering its canonical decomposition into its different eigenfunctions and show empirically that for too small bias the adversarial phenomenon still persists.

References in corpus (13)

• Neural Tangent Kernel: Convergence and Generalization in Neural Networks
• Deep Neural Networks as Gaussian Processes
• On Exact Computation with an Infinitely Wide Neural Net
• A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks
• Nearly-tight VC-dimension and pseudodimension bounds for piecewise linear neural networks
• Graph Neural Tangent Kernel: Fusing Graph Neural Networks with Graph Kernels
• The Convergence Rate of Neural Networks for Learned Functions of Different Frequencies
• Uniform convergence may be unable to explain generalization in deep learning
• Non-Vacuous Generalization Bounds at the ImageNet Scale: A PAC-Bayesian Compression Approach
• Norm-Based Capacity Control in Neural Networks
• Deterministic PAC-Bayesian generalization bounds for deep networks via generalizing noise-resilience
• On the Similarity between the Laplace and Neural Tangent Kernels
• Label-Aware Neural Tangent Kernel: Toward Better Generalization and Local Elasticity

Cited by in corpus (1)

• What can linearized neural networks actually say about generalization?