ADANA thread: horizon-specific momentum tuning can't explain ADANA's edge over AdamW

_katieeverett · x · 2026-09-10

First thread question: is a momentum schedule actually needed? Longer horizons favor longer memory with smaller learning rates, and horizon-specific tuning of momentum constants improves the baselines — but it doesn't explain ADANA's advantage over AdamW.

Related event: Momentum-Scheduled ADANA Changes Scaling Exponents on the Overtraining Axis(13 posts)→

Original post →

More from Research

Research channel →