A Probabilistically Motivated Learning Rate Adaptation for Stochastic Optimization

TL;DR

This paper introduces a probabilistic framework for automatically adapting learning rates in stochastic optimization, improving robustness and reducing manual tuning in deep learning training.

Contribution

It provides a Gaussian inference-based motivation for learning rate adaptation, leading to a meta-algorithm that adjusts learning rates automatically during training.

Findings

Robust adaptation of learning rates across various initial values

Effective in deep learning benchmark tasks

Relates learning rate to a dimensionless, controllable quantity

Abstract

Machine learning practitioners invest significant manual and computational resources in finding suitable learning rates for optimization algorithms. We provide a probabilistic motivation, in terms of Gaussian inference, for popular stochastic first-order methods. As an important special case, it recovers the Polyak step with a general metric. The inference allows us to relate the learning rate to a dimensionless quantity that can be automatically adapted during training by a control algorithm. The resulting meta-algorithm is shown to adapt learning rates in a robust manner across a large range of initial values when applied to deep learning benchmark problems.