End-to-End Differentiable Diffusion Conditioning for Physics-Informed Optimization

The paper introduces Differentiable Diffusion Conditioning (D2C), an offline model‑based optimization approach that turns a frozen conditional diffusion model into a gradient‑based optimizer. By making the reverse denoising chain differentiable with respect to conditioning signals, D2C optimizes conditioning variables using gradients from surrogate objectives and physics‑informed penalties. The method is evaluated on laser pulse optimization for inertial confinement fusion, sustainable data‑center workload scheduling, and four Design‑Bench tasks, achieving top results on the real‑world tasks and the best average rank across the benchmarks.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv AI.

• diffusion