AI Paper Atlas

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

Boosting Multi-Domain Reasoning of LLMs via Curvature-Guided Policy Optimization

Xize Liang, Lin Yang, Jie Wang, Rui Liu, Yang Lu, Jinliang Zeng, Hanzhu Chen, Dong Li, Jianye HAO

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 a curvature-guided policy optimization approach for multi-domain RL of LLMs to mitigate cross-domain conflicts and improve reasoning across domains.

Problem

Multi-domain RL for LLMs has intricate reward surfaces and cross-domain conflicts, where gains in one capability can degrade another.

Approach

Uses curvature information of the reward surface to guide policy optimization so that parameters can excel across domains; the abstract states the goal but the specific mechanism is not described in the provided text.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use CGPO instead of joint learningQwen2.5-7B-Instruct · Multi-domain average (7 benchmarks)average score: 56.62 -> 59.59Improved
use CGPO instead of joint learningQwen2.5-3B-Instruct · Multi-domain average (7 benchmarks)average score: 49.04 -> 50.42Improved
use CGPO instead of joint learningQwen2.5-3B-Instruct · MATH500score: 64.50 -> 64.20No clear effect
use CGPO instead of joint learningQwen2.5-3B-Instruct · WritingBenchscore: 58.61 -> 63.04Improved
use CGPO instead of joint learningQwen2.5-7B-Instruct · MATH500score: 76.00 -> 75.55Hurt
randomize domain order each update (vs fixed order) (ablation)CGPO (Qwen2.5-7B-Instruct) · Multi-domain average (7 benchmarks)average score: 58.48 -> 59.59Improved
set mixing coefficient alpha to 1.2 rather than 0.9 or 1.5 (ablation)CGPO (Qwen2.5-7B-Instruct) · Multi-domain average (7 benchmarks)average score: 58.15 / 58.04 -> 59.59Improved
add CGPO overhead vs joint learningQwen2.5-7B-Instruct · Training timehours per epoch: 17.8 -> 18.6No clear effect

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