Understanding the unstable convergence of gradient descent

Abstract

Most existing analyses of (stochastic) gradient descent rely on the condition that for $L$-smooth costs, the step size is less than $2/L$. However, many works have observed that in machine learning applications step sizes often do not fulfill this condition, yet (stochastic) gradient descent still converges, albeit in an unstable manner. We investigate this unstable convergence phenomenon from first principles, and discuss key causes behind it. We also identify its main characteristics, and how they interrelate based on both theory and experiments, offering a principled view toward understanding the phenomenon.

Accepted to the 39th International Conference on Machine Learning (ICML 2022), Baltimore, Maryland, USA. Version 2 improves writing and presentation, adds discussion regarding concurrent works

Cited by in corpus (1)

• Good-looking but Lacking Faithfulness: Understanding Local Explanation Methods through Trend-based Testing