pith. the verified trust layer for science. sign in

Expected return causes outcome-level mode collapse in re- inforcement learning and how to fix it with inverse probability scaling.CoRR, abs/2601.21669

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it

citation-role summary

background 1

citation-polarity summary

fields

cs.LG 1

years

2026 1

verdicts

UNVERDICTED 1

roles

background 1

polarities

background 1

representative citing papers

The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping

cs.LG · 2026-04-13 · unverdicted · novelty 6.0

MEDS improves LLM RL performance by up to 4.13 pass@1 and 4.37 pass@128 points by dynamically penalizing rollouts matching prevalent historical error clusters identified via memory-stored representations and density clustering.

citing papers explorer

Showing 1 of 1 citing paper.

  • The Past Is Not Past: Memory-Enhanced Dynamic Reward Shaping cs.LG · 2026-04-13 · unverdicted · none · ref 8

    MEDS improves LLM RL performance by up to 4.13 pass@1 and 4.37 pass@128 points by dynamically penalizing rollouts matching prevalent historical error clusters identified via memory-stored representations and density clustering.