Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling

2026-03-16

AI Summary

This work proposes LOOM-CFM, a novel conditional flow matching (CFM) approach that addresses the inefficiency of existing methods in large-scale settings by extending minibatch optimal transport to cross-batch optimization. Unlike conventional CFM techniques that optimize data-noise pairings only within individual minibatches, LOOM-CFM continuously refines these pairings across batches, significantly improving the trade-off between sampling speed and generation quality. The method enhances both the sampling efficiency of continuous normalizing flows and the effectiveness of distillation-based initialization, while also enabling high-resolution image synthesis. Extensive experiments demonstrate that LOOM-CFM achieves superior speed-quality balance across multiple benchmark datasets, establishing a new state of the art in scalable generative modeling with flow-based methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Conditional Flow Matching (CFM), a simulation-free method for training continuous normalizing flows, provides an efficient alternative to diffusion models for key tasks like image and video generation. The performance of CFM in solving these tasks depends on the way data is coupled with noise. A recent approach uses minibatch optimal transport (OT) to reassign noise-data pairs in each training step to streamline sampling trajectories and thus accelerate inference. However, its optimization is restricted to individual minibatches, limiting its effectiveness on large datasets. To address this shortcoming, we introduce LOOM-CFM (Looking Out Of Minibatch-CFM), a novel method to extend the scope of minibatch OT by preserving and optimizing these assignments across minibatches over training time. Our approach demonstrates consistent improvements in the sampling speed-quality trade-off across multiple datasets. LOOM-CFM also enhances distillation initialization and supports high-resolution synthesis in latent space training.

Problem

• Conditional Flow Matching
• minibatch optimal transport
• flow-based generative models
• inference acceleration
• data-noise coupling

Innovation

• Conditional Flow Matching
• Minibatch Optimal Transport
• LOOM-CFM
• Flow-Based Generative Models
• Latent Space Synthesis