Global Convergence of Gradient Descent for Deep Linear Residual Networks

Abstract

We analyze the global convergence of gradient descent for deep linear residual networks by proposing a new initialization: zero-asymmetric (ZAS) initialization. It is motivated by avoiding stable manifolds of saddle points. We prove that under the ZAS initialization, for an arbitrary target matrix, gradient descent converges to an $\epsilon$-optimal point in $O(L^3\log (1/\epsilon ))$ iterations, which scales polynomially with the network depth $L$. Our result and the $\exp (\Omega (L))$ convergence time for the standard initialization (Xavier or near-identity) [Shamir, 2018] together demonstrate the importance of the residual structure and the initialization in the optimization for deep linear neural networks, especially when $L$ is large.

References in corpus (4)

• Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
• Stochastic Gradient Descent Optimizes Over-parameterized Deep ReLU Networks
• Width Provably Matters in Optimization for Deep Linear Neural Networks
• Analysis of the Gradient Descent Algorithm for a Deep Neural Network Model with Skip-connections