SGD at the Edge of Stability: Stochastic Stabilization with Large Learning Rates
For deep learning practitioners, it offers theoretical justification for using large learning rates with SGD, explaining the edge-of-stability phenomenon in stochastic settings.
This work provides convergence guarantees for SGD with large learning rates on multiclass cross-entropy loss, showing that SGD self-stabilizes dynamics and converges in the best-iterate sense despite alternating between edge-of-stability and stable regimes.
Modern deep learning has been shown to operate at the edge of stability, routinely using learning rates far larger than those justified by classical optimization theory. Most prior analyses of the edge of stability phenomenon focus on deterministic gradient descent, leaving the stochastic setting largely unexplored. In this work, we provide sharp convergence guarantees for Stochastic Gradient Descent (SGD) applied to the multiclass cross-entropy loss, for both linear classifiers and two-layer neural networks. We show that the stochasticity of SGD may cause the dynamics to alternate between an edge-of-stability regime that is dominated by curvature-driven oscillations, and a stable regime in which the expected loss decreases at a controlled rate. Despite that, we prove that SGD self-stabilizes the dynamics, ensuring that the iterates return to stability in a fixed number of iterations and allowing convergence in the best-iterate sense even with large learning rates. Experiments validate our theoretical findings and illustrate the benefits of SGD in the large-stepsize regime.