On The Hidden Biases of Flow Matching Samplers

2025-12-18

AI Summary

This work identifies inherent energy and structural biases in empirical flow matching (FM) samplers: their empirical minimizers almost surely fail to constitute gradient fields, resulting in suboptimal velocity field energy. We provide the first rigorous proof of this non-gradient property and establish a novel theoretical framework characterizing source-distribution–driven kinetic energy decay—exponential concentration (both instantaneous and integrated) for Gaussian sources, and polynomial decay for heavy-tailed sources; crucially, this behavior is governed by the source distribution, not the target data. Our methodology integrates empirical FM modeling, optimal transport analysis, kinetic energy concentration inequalities, and probabilistic tail characterization. These results yield the first quantitative characterization of bias in FM samplers and reveal the fundamental role of source distribution selection in shaping sampling dynamics.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds
• Reasoning under Uncertainty: Stochastic Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• Security and Privacy: Large-scale security measurements

Abstract

We study the implicit bias of flow matching (FM) samplers via the lens of empirical flow matching. Although population FM may produce gradient-field velocities resembling optimal transport (OT), we show that the empirical FM minimizer is almost never a gradient field, even when each conditional flow is. Consequently, empirical FM is intrinsically energetically suboptimal. In view of this, we analyze the kinetic energy of generated samples. With Gaussian sources, both instantaneous and integrated kinetic energies exhibit exponential concentration, while heavy-tailed sources lead to polynomial tails. These behaviors are governed primarily by the choice of source distribution rather than the data. Overall, these notes provide a concise mathematical account of the structural and energetic biases arising in empirical FM.

Problem

• Analyzes implicit biases in flow matching samplers
• Shows empirical flow matching is energetically suboptimal
• Examines kinetic energy concentration in generated samples

Innovation

• Empirical flow matching minimizes non-gradient fields
• Kinetic energy concentrates exponentially with Gaussian sources
• Source distribution primarily governs energetic biases