Hausdorff Dimension, Heavy Tails, and Generalization in Neural Networks

Abstract

Despite its success in a wide range of applications, characterizing the generalization properties of stochastic gradient descent (SGD) in non-convex deep learning problems is still an important challenge. While modeling the trajectories of SGD via stochastic differential equations (SDE) under heavy-tailed gradient noise has recently shed light over several peculiar characteristics of SGD, a rigorous treatment of the generalization properties of such SDEs in a learning theoretical framework is still missing. Aiming to bridge this gap, in this paper, we prove generalization bounds for SGD under the assumption that its trajectories can be well-approximated by a \emph{Feller process}, which defines a rich class of Markov processes that include several recent SDE representations (both Brownian or heavy-tailed) as its special case. We show that the generalization error can be controlled by the \emph{Hausdorff dimension} of the trajectories, which is intimately linked to the tail behavior of the driving process. Our results imply that heavier-tailed processes should achieve better generalization; hence, the tail-index of the process can be used as a notion of "capacity metric". We support our theory with experiments on deep neural networks illustrating that the proposed capacity metric accurately estimates the generalization error, and it does not necessarily grow with the number of parameters unlike the existing capacity metrics in the literature.

arXiv commentPublished at NeurIPS 2020 (Spotlight) -- an imprecision in Definition 2 and a mistake in the statement and the proof of Theorem 2 are fixed

References in corpus (27)

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Exploring Generalization in Deep Learning
• Three Factors Influencing Minima in SGD
• Information-theoretic analysis of generalization capability of learning algorithms
• Norm-Based Capacity Control in Neural Networks
• Cooling down Levy flights
• The Landscape of Empirical Risk for Non-convex Losses
• A Variational Analysis of Stochastic Gradient Algorithms
• Generalization Bounds of SGLD for Non-convex Learning: Two Theoretical Viewpoints
• The Heavy-Tail Phenomenon in SGD
• Multiplicative noise and heavy tails in stochastic optimization
• Global Convergence of Stochastic Gradient Hamiltonian Monte Carlo for Non-Convex Stochastic Optimization: Non-Asymptotic Performance Bounds and Momentum-Based Acceleration
• On the diffusion approximation of nonconvex stochastic gradient descent
• Non-Gaussianity of Stochastic Gradient Noise
• Fractional Underdamped Langevin Dynamics: Retargeting SGD with Momentum under Heavy-Tailed Gradient Noise
• On the Heavy-Tailed Theory of Stochastic Gradient Descent for Deep Neural Networks
• On the Convergence of Langevin Monte Carlo: The Interplay between Tail Growth and Smoothness
• Theoretical Issues in Deep Networks: Approximation, Optimization and Generalization
• Is Deeper Better only when Shallow is Good?
• Can SGD Learn Recurrent Neural Networks with Provable Generalization?
• First Exit Time Analysis of Stochastic Gradient Descent Under Heavy-Tailed Gradient Noise
• Multifractal properties of sample paths of ground state-transformed jump processes
• Fractal Structure and Generalization Properties of Stochastic Optimization Algorithms
• Stable behaviour of infinitely wide deep neural networks
• Relativistic Monte Carlo
• The Hausdorff dimension of the range of the Lévy multistable processes
• Asymmetric Heavy Tails and Implicit Bias in Gaussian Noise Injections

Cited by in corpus (10)

• Heavy Tails in SGD and Compressibility of Overparametrized Neural Networks
• Strength of Minibatch Noise in SGD
• Intrinsic Dimension, Persistent Homology and Generalization in Neural Networks
• Rate-Distortion Theoretic Bounds on Generalization Error for Distributed Learning
• Noise and Fluctuation of Finite Learning Rate Stochastic Gradient Descent
• A Fully Spiking Hybrid Neural Network for Energy-Efficient Object Detection
• Heavy-Tail Phenomenon in Decentralized SGD
• Complexity of Quantum Trajectories
• On the Sample Complexity and Metastability of Heavy-tailed Policy Search in Continuous Control
• Approximate Heavy Tails in Offline (Multi-Pass) Stochastic Gradient Descent