judge-efficiency

Tag

Cards List
#judge-efficiency

Rubric Rewards from Item Response Theory

Hugging Face Daily Papers ↗ · 2026-09-28 Cached

The paper introduces Rubric Response Theory (RRT), which replaces additive rubric scoring with a two-parameter item response model plus a Response Parameter Network (RPN) updated via online EM, achieving higher macro criterion scores than GRPO on Medical and Science benchmarks while halving judge requests.

0 favorites 0 likes
← Back to home

Submit Feedback