Relative Fisher Information and Natural Gradient for Large Models

FrnkNlsn · x · 2026-09-01

Highlights a 2017 ICML paper by Frank Nielsen et al. on Relative Fisher Information (RFIM) and relative natural gradient. The method extracts local components from large neural networks to define invariant metrics, simplifying geometric structures to improve the optimization of complex, modular models.

Original post →

More from Research

Research channel →