Smart-GRPO: Smartly Sampling Noise for Efficient RL of Flow-Matching Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yu, Benjamin, Liu, Jackie, Cui, Justin
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918153645195264
author Yu, Benjamin
Liu, Jackie
Cui, Justin
author_facet Yu, Benjamin
Liu, Jackie
Cui, Justin
contents Recent advancements in flow-matching have enabled high-quality text-to-image generation. However, the deterministic nature of flow-matching models makes them poorly suited for reinforcement learning, a key tool for improving image quality and human alignment. Prior work has introduced stochasticity by perturbing latents with random noise, but such perturbations are inefficient and unstable. We propose Smart-GRPO, the first method to optimize noise perturbations for reinforcement learning in flow-matching models. Smart-GRPO employs an iterative search strategy that decodes candidate perturbations, evaluates them with a reward function, and refines the noise distribution toward higher-reward regions. Experiments demonstrate that Smart-GRPO improves both reward optimization and visual quality compared to baseline methods. Our results suggest a practical path toward reinforcement learning in flow-matching frameworks, bridging the gap between efficient training and human-aligned generation.
format Preprint
id arxiv_https___arxiv_org_abs_2510_02654
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Smart-GRPO: Smartly Sampling Noise for Efficient RL of Flow-Matching Models
Yu, Benjamin
Liu, Jackie
Cui, Justin
Computer Vision and Pattern Recognition
Recent advancements in flow-matching have enabled high-quality text-to-image generation. However, the deterministic nature of flow-matching models makes them poorly suited for reinforcement learning, a key tool for improving image quality and human alignment. Prior work has introduced stochasticity by perturbing latents with random noise, but such perturbations are inefficient and unstable. We propose Smart-GRPO, the first method to optimize noise perturbations for reinforcement learning in flow-matching models. Smart-GRPO employs an iterative search strategy that decodes candidate perturbations, evaluates them with a reward function, and refines the noise distribution toward higher-reward regions. Experiments demonstrate that Smart-GRPO improves both reward optimization and visual quality compared to baseline methods. Our results suggest a practical path toward reinforcement learning in flow-matching frameworks, bridging the gap between efficient training and human-aligned generation.
title Smart-GRPO: Smartly Sampling Noise for Efficient RL of Flow-Matching Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.02654