From Imitation to Discrimination: Toward A Generalized Curriculum Advantage Mechanism Enhancing Cross-Domain Reasoning Tasks

TL;DR

This paper introduces CAPO, an adaptive curriculum advantage mechanism that enhances cross-domain reasoning in language models by effectively utilizing advantage signals for improved generalization and robustness.

Contribution

The paper proposes CAPO, a novel curriculum advantage mechanism that leverages advantage signals to improve reasoning and generalization in language models across multiple domains.

Findings

Consistent improvements in mathematical reasoning tasks.

Effective generalization to multimodal GUI reasoning scenarios.

Compatible with diverse optimization methods.

Abstract

Reinforcement learning has emerged as a paradigm for post-training large language models, boosting their reasoning capabilities. Such approaches compute an advantage value for each sample, reflecting better or worse performance than expected, thereby yielding both positive and negative signals for training. However, the indiscriminate mixing of the two signals in existing methods, especially from the early stages, may lead to ambiguous guidance and limited gains. To address this issue, we propose **CAPO** (**C**urriculum **A**dvantage **P**olicy **O**ptimization), an adaptive curriculum mechanism based on advantage signals. The proposed mechanism bootstraps imitation learning with positive-only advantage samples to establish robust foundations, and subsequently introduces negative signals to cultivate discriminative capabilities, thereby improving generalization across complex scenarios. Compatible with diverse optimization methods including GRPO, PPO, RLOO, and Reinforce++, our method consistently achieves stable and significant improvements in mathematical reasoning tasks, and further generalizes effectively to multimodal Graphical User Interface (GUI) reasoning scenarios, establishing itself as a versatile and robust optimization framework.