Alternating Reinforcement Learning with Rubric Rewards (ARL-RR) replaces brittle scalar reward aggregation with a semantic meta-class optimization framework.
March 18, 2026
Original Paper
Alternating Reinforcement Learning with Contextual Rubric Rewards
arXiv · 2603.15646
The Takeaway
It moves alignment away from sensitive scalar weights toward dynamic multi-dimensional rubric optimization, showing consistent improvements in training efficiency and model performance across scales up to 14B.
From the abstract
Reinforcement Learning with Rubric Rewards (RLRR) is a framework that extends conventional reinforcement learning from human feedback (RLHF) and verifiable rewards (RLVR) by replacing scalar preference signals with structured, multi-dimensional, contextual rubric-based evaluations. However, existing approaches in RLRR are limited to linearly compressing vector rewards into a scalar reward with a fixed weightings, which is sensitive to artificial score design and fails to capture correlations amo