CC-LEARN: Cohort-based Consistency Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ye, Xiao, Shrivastava, Shaswat, Li, Zhaonan, Dineen, Jacob, Lu, Shijie, Ahuja, Avneet, Shen, Ming, Xu, Zhikun, Zhou, Ben
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909652619362304
author Ye, Xiao
Shrivastava, Shaswat
Li, Zhaonan
Dineen, Jacob
Lu, Shijie
Ahuja, Avneet
Shen, Ming
Xu, Zhikun
Zhou, Ben
author_facet Ye, Xiao
Shrivastava, Shaswat
Li, Zhaonan
Dineen, Jacob
Lu, Shijie
Ahuja, Avneet
Shen, Ming
Xu, Zhikun
Zhou, Ben
contents Large language models excel at many tasks but still struggle with consistent, robust reasoning. We introduce Cohort-based Consistency Learning (CC-Learn), a reinforcement learning framework that improves the reliability of LLM reasoning by training on cohorts of similar questions derived from shared programmatic abstractions. To enforce cohort-level consistency, we define a composite objective combining cohort accuracy, a retrieval bonus for effective problem decomposition, and a rejection penalty for trivial or invalid lookups that reinforcement learning can directly optimize, unlike supervised fine-tuning. Optimizing this reward guides the model to adopt uniform reasoning patterns across all cohort members. Experiments on challenging reasoning benchmarks (including ARC-Challenge and StrategyQA) show that CC-Learn boosts both accuracy and reasoning stability over pretrained and SFT baselines. These results demonstrate that cohort-level RL effectively enhances reasoning consistency in LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2506_15662
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CC-LEARN: Cohort-based Consistency Learning
Ye, Xiao
Shrivastava, Shaswat
Li, Zhaonan
Dineen, Jacob
Lu, Shijie
Ahuja, Avneet
Shen, Ming
Xu, Zhikun
Zhou, Ben
Computation and Language
Large language models excel at many tasks but still struggle with consistent, robust reasoning. We introduce Cohort-based Consistency Learning (CC-Learn), a reinforcement learning framework that improves the reliability of LLM reasoning by training on cohorts of similar questions derived from shared programmatic abstractions. To enforce cohort-level consistency, we define a composite objective combining cohort accuracy, a retrieval bonus for effective problem decomposition, and a rejection penalty for trivial or invalid lookups that reinforcement learning can directly optimize, unlike supervised fine-tuning. Optimizing this reward guides the model to adopt uniform reasoning patterns across all cohort members. Experiments on challenging reasoning benchmarks (including ARC-Challenge and StrategyQA) show that CC-Learn boosts both accuracy and reasoning stability over pretrained and SFT baselines. These results demonstrate that cohort-level RL effectively enhances reasoning consistency in LLMs.
title CC-LEARN: Cohort-based Consistency Learning
topic Computation and Language
url https://arxiv.org/abs/2506.15662