AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

Sample Lottery: Unsupervised Discovery of Critical Instances for LLM Reasoning

Zhiping Xiao, Yusheng Zhao, Qixin Zhang, Jiaye Xie, Wanjia Zhao, Weizhi Zhang, Xiao Luo, Philip S. Yu, Ming Zhang

ICLR 2026 · Large language models › Reasoning and chain of thought · approach: RL with verifiable rewards for reasoning

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes Sample Lottery, an unsupervised method that identifies a small set of critical training instances for RLVR-based LLM reasoning, so that policy optimization need not rely on full annotation or uniform compute over all samples.

Problem

RLVR for LLM reasoning typically requires fully annotated datasets and allocates compute uniformly across all samples, which is inefficient.

Approach

Frames a "lottery sample hypothesis" for policy optimization of LLMs: a large training set contains a small subset of critical instances sufficient for effective RLVR, and discovers them without supervision.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train on 8 CONST-selected questionsLLaMA-3.1-8B-Instruct · Math avgavg@k accuracy: 20.12 -> 28.31Improved
train on 8 CONST-selected instead of 8 random questionsLLaMA-3.1-8B-Instruct · Math avgavg@k accuracy: 22.32 -> 28.31Improved
CONST with 8 annotated questions vs full-dataset annotationLLaMA-3.1-8B-Instruct · Math avgavg@k accuracy: 28.03 -> 28.31No clear effect
CONST with 8 annotated questions vs full datasetQwen2.5-Math-7B · Math avgavg@k accuracy: 55.00 -> 54.94No clear effect
CONST with 8 questions vs 8 random questionsQwen2.5-Math-7B · Math avgavg@k accuracy: 53.52 -> 54.94Improved
CONST vs best active learning baseline (budget 8)DeepSeek-R1-Distill-Qwen-1.5B · Math avgavg@k accuracy: 45.60 -> 48.14Improved
add conformal prediction set size selection (vs random within cluster) (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 23.01 -> 28.31Improved
add conformal prediction set size selection (vs entropy) (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 23.30 -> 28.31Improved
add clustering for diversity (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 25.54 -> 28.31Improved
add procedural volatility (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 26.38 -> 28.31Improved
add outcome volatility (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 23.04 -> 28.31Improved
select 1 question per b clusters vs 2 per b/2 clusters (ablation)CONST · Math avg, budget 8, LLaMA-3.1-8B-Instructavg@k accuracy: 27.60 -> 28.31Improved

Experimental setup

Builds on: RLVR lottery ticket hypothesis

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