AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Beyond Normalization: Rethinking the Partition Function as a Difficulty Scheduler for RLVR

Dohyung Kim, Minbeom Kim, Jeonghye Kim, Sangmook Lee, Sojeong Rhee, Kyomin Jung

ICML 2026 · Reinforcement learning › RL from feedback · approach: Scaling RL with verifiable rewards · 0 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add partition-function-guided prompt selection and replay (PACED-RL) over GRPOQwen2.5-Math-7B · Math avg of five benchmarkspass@1: 37.2 -> 40.1Improved
PACED-RL versus FlowRLQwen2.5-Math-1.5B · Math avg of five benchmarkspass@1: 27.1 -> 32.1Improved
add accuracy-estimation-error-prioritized replay (ablation)PACED-RL (Qwen2.5-Math-1.5B) · Math avg of five benchmarkspass@1: 30.2 -> 32.1Improved
add replay on AIME 24 (ablation)PACED-RL (Qwen2.5-Math-1.5B) · AIME 24 Avg@32pass@1: 9.8 -> 13.3Improved
use greedy top-m selection instead of softmax soft sampling (ablation)PACED-RL (Qwen2.5-Math-1.5B) · MATH-500 Avg@8Avg@8: 72.2 -> 73.2Improved
target accuracy tau=0.5 instead of 0.1 or 0.9 (ablation)PACED-RL (Qwen2.5-Math-1.5B) · MinervaMath test accuracytest accuracyImproved
target accuracy tau=0.1 (ablation)PACED-RL (Qwen2.5-Math-1.5B) · MinervaMath test accuracytest accuracyHurt
omit KL term in accuracy estimator (Eq. 8 vs Eq. 7) (ablation)PACED-RL · prompt selection overlapselection overlapNo clear effect
learn partition function instead of batch-estimate VarGrad (ablation)FlowRL (Qwen2.5-Math-1.5B) · 5 math benchmarkstest performanceImproved
use VarGrad batch-estimate partition function in FlowRL (ablation)FlowRL (Qwen2.5-Math-1.5B) · 5 math benchmarkstest performanceImproved
use partition function accuracy estimates instead of over-sampling (DS/LILO)PACED-RL (Qwen2.5-Math-7B) · Math avg of five benchmarkspass@1 and rollouts: LILO 40.1 vs PACED-RL 40.1; rollouts 4.0x -> 1.0xNo clear effect
over-sampling prompt selection (DS, LILO) for code generationDeepSeek-R1-Distill-Qwen-1.5B · HumanEval+ / LiveCodeBenchpass@1 vs wall-clockHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy