Paper: Mirror Descent and Novel Exponentiated Gradient Algorithms Using Trace-Form Entropies

FrnkNlsn · x · 2026-08-01

A new arXiv paper introduces a broad class of Mirror Descent (MD) and Generalized Exponentiated Gradient (GEG) algorithms derived from trace-form entropies. The research demonstrates that leveraging these generalized entropies improves convergence behavior and enhances robustness against vanishing and exploding gradients. The framework reveals deep connections between additive, multiplicative, and natural gradient updates under a unified geometric perspective.

Related event: New Optimization Algorithms Based on Trace-Form Entropy(2 posts)→

Original post →

More from Research

Research channel →