Categorical Flow Maps

2026-02-12

AI Summary

This work addresses the challenge of efficient generation for discrete categorical data—such as images, molecular graphs, and text—by proposing a simplex-oriented continuous flow matching approach. It introduces continuous flow matching to discrete generative modeling for the first time, leveraging simplex constraints to naturally restrict the prediction space. By integrating self-distillation with an endpoint consistency objective, the method enables high-quality sampling in very few steps, even in a single step. The framework seamlessly accommodates existing guidance and reweighting techniques, facilitating controllable generation. Evaluated across image, molecular graph, and text generation tasks, the approach substantially outperforms current few-step methods and achieves competitive performance even in the single-step regime.

Technology Category

• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

We introduce Categorical Flow Maps, a flow-matching method for accelerated few-step generation of categorical data via self-distillation. Building on recent variational formulations of flow matching and the broader trend towards accelerated inference in diffusion and flow-based models, we define a flow map towards the simplex that transports probability mass toward a predicted endpoint, yielding a parametrisation that naturally constrains model predictions. Since our trajectories are continuous rather than discrete, Categorical Flow Maps can be trained with existing distillation techniques, as well as a new objective based on endpoint consistency. This continuous formulation also automatically unlocks test-time inference: we can directly reuse existing guidance and reweighting techniques in the categorical setting to steer sampling toward downstream objectives. Empirically, we achieve state-of-the-art few-step results on images, molecular graphs, and text, with strong performance even in single-step generation.

Problem

• categorical data
• few-step generation
• flow matching
• accelerated inference
• generative modeling

Innovation

• Categorical Flow Maps
• flow matching
• self-distillation
• continuous trajectory
• few-step generation