Low-Pass Flow Matching
For generative modeling practitioners, this offers a principled way to reduce sampling cost in flow matching without sacrificing quality, though improvements are incremental.
Low-Pass Flow Matching introduces a time-varying spectral bias to align flow matching with natural data's frequency-decaying power spectra, improving sampling efficiency (e.g., fewer ODE steps) while maintaining or improving image quality on Galaxy10 and other datasets.
Flow Matching typically relies on white noise sources, a choice often misaligned with the power spectra of natural data, which tend to decay with frequency. To address this, we introduce Low-Pass Flow Matching, a variant of Flow Matching based on an operator-modulated interpolant. This formulation induces a time-varying spectral bias that transitions from the source spectrum to a frequency-decaying bias as the path approaches the data. We validate our method on unconditional image generation tasks, including the scientific Galaxy10 dataset. Empirically, we show that our method is particularly effective when paired with adaptive ODE solvers, where it improves or preserves sample quality while substantially reducing sampling cost compared to standard baselines.