Derives backward and forward corrections for asymmetric verifier noise that improve RLVR performance on math reasoning tasks.
Tinyv: Reducing false negatives in verification improves rl for llm reasoning
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
representative citing papers
Gated synthetic augmentations of a 10-task human base substitute for ~87 extra human RLVR tasks on aggregate held-out pass@1, with cost-adjusted trade rate ρ_cost in [1.4×, 11.6×].
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
This review synthesizes representative advances in high-dimensional statistics, highlights common themes and open problems, and points to key entry works.
citing papers explorer
-
Reinforcement Learning with Verifiable yet Noisy Rewards under Imperfect Verifiers
Derives backward and forward corrections for asymmetric verifier noise that improve RLVR performance on math reasoning tasks.
-
Trading Human Curation for Synthetic Augmentation in RLVR
Gated synthetic augmentations of a 10-task human base substitute for ~87 extra human RLVR tasks on aggregate held-out pass@1, with cost-adjusted trade rate ρ_cost in [1.4×, 11.6×].
-
Trust Region On-Policy Distillation
TrOPD stabilizes on-policy distillation for LLMs with trust-region learning, outlier estimation, and off-policy guidance, outperforming prior OPD methods on reasoning and code benchmarks.
-
High-Dimensional Statistics: Reflections on Progress and Open Problems
This review synthesizes representative advances in high-dimensional statistics, highlights common themes and open problems, and points to key entry works.
- Delay, Plateau, or Collapse: Evaluating the Impact of Systematic Verification Error on RLVR