Discrete Meanflow Training Curriculum

2026-04-10

AI Summary

Although the Meanflow model demonstrates strong performance in one-step and few-step sampling, its training incurs substantial computational and data costs. This work proposes a Discrete Meanflow (DMF) training framework that, for the first time, systematically incorporates discretization-based consistency properties into Meanflow training, establishing a self-consistent curriculum. By combining fine-tuning of a pretrained flow model with a discretized objective function, the method substantially reduces training overhead. Remarkably, with only 2,000 training iterations, it achieves a one-step sampling FID of 3.36 on CIFAR-10, significantly enhancing both training efficiency and practical applicability.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Learning with Manifolds
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Flow-based image generative models exhibit stable training and produce high quality samples when using multi-step sampling procedures. One-step generative models can produce high quality image samples but can be difficult to optimize as they often exhibit unstable training dynamics. Meanflow models exhibit excellent few-step sampling performance and tantalizing one-step sampling performance. Notably, MeanFlow models that achieve this have required extremely large training budgets. We significantly decrease the amount of computation and data budget it takes to train Meanflow models by noting and exploiting a particular discretization of the Meanflow objective that yields a consistency property which we formulate into a ``Discrete Meanflow''(DMF) Training Curriculum. Initialized with a pretrained Flow Model, DMF curriculum reaches one-step FID 3.36 on CIFAR-10 in only 2000 epochs. We anticipate that faster training curriculums of Meanflow models, specifically those fine-tuned from existing Flow Models, drives efficient training methods of future one-step examples.

Problem

• Meanflow
• one-step generation
• training efficiency
• flow-based models
• discretization

Innovation

• Discrete Meanflow
• Training Curriculum
• One-step Sampling
• Flow-based Generative Models
• Consistency Property