CMAD: Cooperative Multi-Agent Diffusion via Stochastic Optimal Control

TL;DR

This paper introduces a novel framework for composing multiple pre-trained diffusion models as cooperative agents using stochastic optimal control, enabling more flexible and effective generative tasks.

Contribution

It formulates multi-model composition as a cooperative stochastic optimal control problem, moving beyond traditional algebraic density combination methods.

Findings

Validated on conditional MNIST generation.

Outperforms naive gradient guidance baseline.

Demonstrates effective cooperative control of diffusion models.

Abstract

Continuous-time generative models have achieved remarkable success in image restoration and synthesis. However, controlling the composition of multiple pre-trained models remains an open challenge. Current approaches largely treat composition as an algebraic composition of probability densities, such as via products or mixtures of experts. This perspective assumes the target distribution is known explicitly, which is almost never the case. In this work, we propose a different paradigm that formulates compositional generation as a cooperative Stochastic Optimal Control problem. Rather than combining probability densities, we treat pre-trained diffusion models as interacting agents whose diffusion trajectories are jointly steered, via optimal control, toward a shared objective defined on their aggregated output. We validate our framework on conditional MNIST generation and compare it against a na\"ive inference-time DPS-style baseline replacing learned cooperative control with per-step gradient guidance.