Stronger Convergence Results for Deep Residual Networks: Network Width Scales Linearly with Training Data Size

TL;DR

This paper proves that deep residual networks with width scaling linearly with training data size can reliably converge to global minima using gradient descent, regardless of network depth, under certain conditions.

Contribution

It provides the first theoretical guarantee for convergence of ResNets when network width scales linearly with data size, independent of depth.

Findings

Gradient descent converges to global minima when Jacobian is full rank.

Network width can scale linearly with data size for guaranteed convergence.

Convergence results are independent of network depth.

Abstract

Deep neural networks are highly expressive machine learning models with the ability to interpolate arbitrary datasets. Deep nets are typically optimized via first-order methods and the optimization process crucially depends on the characteristics of the network as well as the dataset. This work sheds light on the relation between the network size and the properties of the dataset with an emphasis on deep residual networks (ResNets). Our contribution is that if the network Jacobian is full rank, gradient descent for the quadratic loss and smooth activation converges to the global minima even if the network width $m$ of the ResNet scales linearly with the sample size $n$, and independently from the network depth. To the best of our knowledge, this is the first work which provides a theoretical guarantee for the convergence of neural networks in the $m=\Omega(n)$ regime.