Guidance Contrastive Token Credit Assignment for Discrete Policy Optimization

2026-05-27

AI Summary

This work addresses the coarse credit assignment inherent in existing sample-level reward-based reinforcement learning methods for discrete policy optimization, which struggle to discern fine-grained contributions of individual tokens. To overcome this limitation, the paper proposes Guided Contrastive Policy Optimization (GCPO), a novel algorithm that introduces, for the first time, a prediction contrast mechanism guided by positive and negative prompts to enable token-level advantage estimation and credit assignment. By integrating contrastive learning with discrete policy gradients, GCPO generates more precise learning signals. Experimental results demonstrate that GCPO significantly outperforms baseline methods such as GRPO and DAPO on both text-to-image generation and chain-of-thought reasoning tasks, confirming its effectiveness and broad applicability.

Abstract

Group-advantage-based reinforcement learning methods, such as GRPO and DAPO, have demonstrated strong performance across diverse domains, including mathematical reasoning and text-to-image generation. However, their reliance on sample-level rewards introduces a key limitation as uniform credit assignment across all tokens fails to capture fine-grained, token-level contributions. To address this issue, we propose Guidance Contrastive Policy Optimization (GCPO), a novel algorithm that enables per-token credit assignment by contrasting model predictions under positive and negative prompts. Rather than uniformly broadcasting sample-level advantages, GCPO assigns token-level advantages proportional to the difference between these contrastive predictions, allowing more precise and informative learning signals. Empirically, we find that GCPO emphasizes semantically relevant regions such as visual areas aligned with textual prompts in text-to-image generation, and critical keywords within reasoning traces for chain-of-thought tasks. Through extensive experiments, GCPO consistently outperforms GRPO and DAPO baselines on both text-to-image generation and chain-of-thought reasoning benchmarks, demonstrating its effectiveness as a general and scalable optimization strategy for discrete policy learning.

Problem

• credit assignment
• discrete policy optimization
• token-level reward
• reinforcement learning
• fine-grained contribution

Innovation

• token-level credit assignment
• contrastive policy optimization
• discrete policy learning
• guidance contrast
• reinforcement learning