ReflexFlow: Rethinking Learning Objective for Exposure Bias Alleviation in Flow Matching

2025-12-04

AI Summary

Flow Matching suffers from training-inference inconsistency, leading to exposure bias: models struggle to generalize to deviated inputs, and insufficient low-frequency information recovery in early denoising stages causes error accumulation. This paper is the first to systematically revisit this issue at the training-objective level and proposes a model-agnostic dynamic correction framework comprising two complementary modules—anti-drift correction and frequency compensation—to enable self-referential correction of deviated inputs and enhanced low-frequency content reconstruction. The method introduces a scheduling-aware redesigned loss, exposure-bias-aware loss reweighting, and a frequency-adaptive compensation mechanism. Experiments on CIFAR-10, CelebA-64, and ImageNet-256 demonstrate substantial improvements in generation quality; notably, FID on CelebA-64 drops by 35.65%, outperforming existing Flow Matching approaches.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Despite tremendous recent progress, Flow Matching methods still suffer from exposure bias due to discrepancies in training and inference. This paper investigates the root causes of exposure bias in Flow Matching, including: (1) the model lacks generalization to biased inputs during training, and (2) insufficient low-frequency content captured during early denoising, leading to accumulated bias. Based on these insights, we propose ReflexFlow, a simple and effective reflexive refinement of the Flow Matching learning objective that dynamically corrects exposure bias. ReflexFlow consists of two components: (1) Anti-Drift Rectification (ADR), which reflexively adjusts prediction targets for biased inputs utilizing a redesigned loss under training-time scheduled sampling; and (2) Frequency Compensation (FC), which reflects on missing low-frequency components and compensates them by reweighting the loss using exposure bias. ReflexFlow is model-agnostic, compatible with all Flow Matching frameworks, and improves generation quality across datasets. Experiments on CIFAR-10, CelebA-64, and ImageNet-256 show that ReflexFlow outperforms prior approaches in mitigating exposure bias, achieving a 35.65% reduction in FID on CelebA-64.

Problem

• Addresses exposure bias in Flow Matching methods
• Proposes ReflexFlow for dynamic correction of bias
• Enhances generation quality across multiple datasets

Innovation

• ReflexFlow dynamically corrects exposure bias in Flow Matching
• Anti-Drift Rectification adjusts prediction targets for biased inputs
• Frequency Compensation reweights loss to recover low-frequency content