CMAD: Cooperative Multi-Agent Diffusion via Stochastic Optimal Control
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913145790922752 |
|---|---|
| author | Barbano, Riccardo Denker, Alexander Kereta, Zeljko Li, Runchang Vargas, Francisco |
| author_facet | Barbano, Riccardo Denker, Alexander Kereta, Zeljko Li, Runchang Vargas, Francisco |
| contents | 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 naïve inference-time DPS-style baseline replacing learned cooperative control with per-step gradient guidance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_10933 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CMAD: Cooperative Multi-Agent Diffusion via Stochastic Optimal Control Barbano, Riccardo Denker, Alexander Kereta, Zeljko Li, Runchang Vargas, Francisco Machine Learning 93E20 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 naïve inference-time DPS-style baseline replacing learned cooperative control with per-step gradient guidance. |
| title | CMAD: Cooperative Multi-Agent Diffusion via Stochastic Optimal Control |
| topic | Machine Learning 93E20 |
| url | https://arxiv.org/abs/2602.10933 |