trends.barnas.net

method · first seen 5 Nov 2026

group-relative baseline subtraction

Also: group-mean baseline

Variance-reducing baseline in policy gradients computed as the mean reward over a group of sampled responses.

Mentions per week

0119 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

low-rank adaptationpost-training