arXiv cs.AI· Ricardo Luna Gutierrez, Vineet Gundecha, Rahman Ejaz, Varchas Gopalaswamy, Riccardo Betti, Sahand Ghorbanpour, Aarne Lees, Soumyendu Sarkar·· 12 小时前AI 评分24
D2C：面向物理约束优化的端到端可微扩散条件化方法
End-to-End Differentiable Diffusion Conditioning for Physics-Informed Optimization
AI 导读
研究者提出 Differentiable Diffusion Conditioning（D2C），利用确定性采样使逆向去噪链对条件信号可微，将冻结的条件扩散模型转为端到端梯度优化器，在测试时通过冻结代理目标与物理约束惩罚的梯度直接优化条件变量。该方法在惯性约束聚变激光脉冲优化和数据中心负载调度两个真实任务上取得最强结果，并在四个 Design-Bench 任务上取得最佳平均排名。
正文 · 原文
Abstract:Optimizing high-dimensional physical systems under expensive evaluations and complex constraints is a central challenge in science and engineering. In offline model-based optimization (MBO), methods must improve designs using only a fixed dataset and no test-time access to the true objective. However, existing methods typically suffer from conservatism, unconstrained surrogate exploitation, or function as static inference-time samplers that cannot refine candidates against precise physical objectives. In this paper, we propose Differentiable Diffusion Conditioning (D2C) for offline MBO, which leverages the fact that deterministic sampling makes the reverse denoising chain differentiable with respect to the conditioning signal. D2C turns a frozen conditional diffusion model into an end-to-end gradient-based optimizer by directly optimizing conditioning variables at test time via gradients from frozen surrogate objectives and physics-informed penalties backpropagated through the reverse chain. We implement this idea in a composed proposer-evaluator framework and evaluate it on laser pulse optimization for inertial confinement fusion, sustainable data-center workload scheduling, and four Design-Bench tasks. Our method achieves the strongest results on the two real-world tasks and the best average rank across four Design-Bench tasks.
| Comments: | Initial version submitted to the NeurIPS 2026 main conference pending publication |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.31721 [cs.LG] |
| (or arXiv:2610.31721v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.31721 arXiv-issued DOI via DataCite |
Submission history
From: Soumyendu Sarkar [view email]
[v1]
Thu, 22 Oct 2026 09:53:05 UTC (209 KB)
来源：arXiv cs.AI · arxiv.org