Research map / Image generation / Fast diffusion sampling
End-to-End Differentiable Diffusion Conditioning for Physics-Informed Optimization
Ricardo Luna Gutierrez, Vineet Gundecha, Rahman Ejaz, Varchas Gopalaswamy, Riccardo Betti, Sahand Ghorbanpour, Aarne Lees, Soumyendu Sarkar
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes an end-to-end differentiable diffusion conditioning method for offline model-based optimization of physical systems, as described in the truncated abstract.
Problem
Offline model-based optimization of high-dimensional physical systems, where designs must be improved using only a fixed dataset and no test-time access to the true objective; existing methods suffer from conservatism, unconstrained surrogate exploitation, or act as static inference.
Approach
Uses a diffusion model conditioned end-to-end through a differentiable pipeline for physics-informed optimization; the abstract is truncated so the specific mechanism is not fully stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| optimize the conditioning with gradients through the DDIM chain instead of a forward-only conditioning sweep (Brute Force) | frozen diffusion proposer · ICF Top | neutron yield improvement (%): 3.93 -> 10.06 | Improved |
| optimize the conditioning with gradients through the DDIM chain instead of a forward-only conditioning sweep (Brute Force) | frozen diffusion proposer · ICF Standard | neutron yield improvement (%): 4.66 -> 17.98 | Improved |
| use D2C instead of the best generative baseline GTG | data center scheduling · Data center scheduling | carbon footprint reduction (%): 5.50 -> 8.98 | Improved |
| optimize conditioning instead of latent noise (real-world tasks) (ablation) | D2C · ICF and DC | constraint-satisfying gain | Improved |
| optimize initial latent noise (conditioning fixed to c=1.0) instead of the 1-D conditioning (ablation) | D2C · Design-Bench | normalized score | Improved |
| D2C versus DDOM on Ant Morphology | D2C · Design-Bench Ant | normalized score: 0.925 (DDOM) vs 0.876 | Hurt |
| D2C versus GTG on D'Kitty | D2C · Design-Bench D'Kitty | normalized score: 0.976 (GTG) vs 0.974 | No clear effect |
| run D2C on expert-designed pulses with direct LILAC simulation | D2C · ICF expert pulses (20) | average yield gain (%): +9.8% | Improved |
Limitations
- Full text unavailable; abstract truncated, so details are limited
Closest papers by meaning
- Arbitrarily-Conditioned Multi-Functional Diffusion for Multi-Physics Emulation (ICML 2025 · similarity 0.89)
- Mosaic: A Benchmark Suite for Differentiable Physics Solvers (NeurIPS 2026 · similarity 0.86)
- Iterated Denoising Energy Matching for Sampling from Boltzmann Densities (ICML 2024 · similarity 0.85)
- Masked Diffusion Models as Energy Minimization (NeurIPS 2025 · similarity 0.85)
- Enhanced Diffusion Sampling via Extrapolation with Multiple ODE Solutions (ICLR 2025 · similarity 0.85)
- Physics-Informed Diffusion Models (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
