Non-Singularity of the Gradient Descent map for Neural Networks with Piecewise Analytic Activations

TL;DR

This paper proves that the gradient descent map is non-singular for realistic neural networks with piecewise analytic activations, which supports theoretical guarantees for avoiding saddle points and convergence to global minima.

Contribution

It establishes the non-singularity of the GD map for practical neural network architectures with common activation functions, extending prior theoretical results.

Findings

GD map is non-singular for almost all step-sizes in realistic networks

Supports theoretical guarantees for avoiding saddle points

Extends convergence analysis to practical neural network settings

Abstract

The theory of training deep networks has become a central question of modern machine learning and has inspired many practical advancements. In particular, the gradient descent (GD) optimization algorithm has been extensively studied in recent years. A key assumption about GD has appeared in several recent works: the \emph{GD map is non-singular} -- it preserves sets of measure zero under preimages. Crucially, this assumption has been used to prove that GD avoids saddle points and maxima, and to establish the existence of a computable quantity that determines the convergence to global minima (both for GD and stochastic GD). However, the current literature either assumes the non-singularity of the GD map or imposes restrictive assumptions, such as Lipschitz smoothness of the loss (for example, Lipschitzness does not hold for deep ReLU networks with the cross-entropy loss) and restricts the analysis to GD with small step-sizes. In this paper, we investigate the neural network map as a function on the space of weights and biases. We also prove, for the first time, the non-singularity of the gradient descent (GD) map on the loss landscape of realistic neural network architectures (with fully connected, convolutional, or softmax attention layers) and piecewise analytic activations (which includes sigmoid, ReLU, leaky ReLU, etc.) for almost all step-sizes. Our work significantly extends the existing results on the convergence of GD and SGD by guaranteeing that they apply to practical neural network settings and has the potential to unlock further exploration of learning dynamics.