AI Paper Atlas

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

One-shot Entropy Minimization for Language Model Reasoning

Zitian Gao, Yilong Chen, Haoming Luo, Joey Zhou, Bryan Dai

ICML 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling

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 One-shot Entropy Minimization, a fully unsupervised post-training method that minimizes token entropy on a single unlabeled example for about ten gradient steps to improve LM reasoning.

Problem

Improving language model reasoning and generation without labels, and avoiding data contamination when evaluating unsupervised post-training gains.

Approach

Applies entropy minimization using only a single unlabeled example and roughly ten gradient steps, and pretrains a 7B model from scratch on strictly decontaminated data to isolate the effect from pretraining contamination.

Key results

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add 1-shot entropy minimizationQwen2.5-Math-7B · 6-benchmark averageaverage score: 29.2 -> 51.1Improved
add 1-shot entropy minimizationDeconta-7B-SFT · 6-benchmark averageaverage score: 52.1 -> 56.6Improved
add 1-shot entropy minimization on top of RLDeconta-7B-RL · 6-benchmark averageaverage score: 59.5 -> 62.3Improved
add 1-shot entropy minimizationQwen3-8B-Base · 6-benchmark averageaverage score: 29.7 -> 34.6Improved
add 1-shot entropy minimizationLlama-3.1-Nemotron-Nano-8B-v1 · 6-benchmark averageaverage score: 40.6 -> 44.8Improved
add variance-based data selection (ablation)1-shot EM on Deconta-7B-SFT · 6-benchmark averageaverage score: 55.0 -> 56.6Improved
add variance-based data selection (ablation)1-shot EM on Deconta-7B-RL · 6-benchmark averageaverage score: 61.7 -> 62.3Improved
add variance-based data selection (ablation)1-shot EM on Qwen3-8B-Base · 6-benchmark averageaverage score: 33.9 -> 34.6Improved
add variance-based data selection (ablation)1-shot EM on Qwen2.5-Math-7B · 6-benchmark averageaverage score: 49.5 -> 51.1Improved
add variance-based data selection (ablation)1-shot EM on Llama-3.1-Nemotron-Nano-8B-v1 · 6-benchmark averageaverage score: 44.1 -> 44.8Improved
train EM beyond about 10 steps (ablation)1-shot EM · math reasoningaverage scoreHurt
apply entropy minimization instead of RLlogit distribution (Qwen2.5-Math-7B) · flattened logitsskewness: 0.11 -> 1.54Mixed

Experimental setup

Datasets: single unlabeled example

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