Stepwise Credit Assignment for GRPO on Flow-Matching Models
This addresses a specific bottleneck in reinforcement learning for flow models, offering an incremental improvement over existing methods.
The paper tackles the problem of uniform credit assignment in Flow-GRPO for flow-matching models, which ignores the temporal structure of diffusion generation and can reward suboptimal intermediate steps. The proposed Stepwise-Flow-GRPO method, which assigns credit based on each step's reward improvement using Tweedie's formula and gain-based advantages, achieves superior sample efficiency and faster convergence.
Flow-GRPO successfully applies reinforcement learning to flow models, but uses uniform credit assignment across all steps. This ignores the temporal structure of diffusion generation: early steps determine composition and content (low-frequency structure), while late steps resolve details and textures (high-frequency details). Moreover, assigning uniform credit based solely on the final image can inadvertently reward suboptimal intermediate steps, especially when errors are corrected later in the diffusion trajectory. We propose Stepwise-Flow-GRPO, which assigns credit based on each step's reward improvement. By leveraging Tweedie's formula to obtain intermediate reward estimates and introducing gain-based advantages, our method achieves superior sample efficiency and faster convergence. We also introduce a DDIM-inspired SDE that improves reward quality while preserving stochasticity for policy gradients.