NeurIPS paper steers crystal diffusion models with Monte Carlo tree search
CatAstro_Piyush · x · 2026-11-21
A McGill team presents "Diffusion Tree Sampling for Inference Time Adaptation of Crystal Generative Models" (arXiv:2611.17462) at NeurIPS Sydney. Instead of costly per-objective fine-tuning, DTS adapts diffusion models at inference time via Monte Carlo tree search over the denoising trajectory. Two contributions: existing property-conditional generative models can serve as guided proposals inside DTS without distorting the target distribution, and DTS is extended to multi-objective sampling so one search spans an entire Pareto front. Demonstrated on de novo crystal generation: improved stability, rare space groups, high bulk modulus, and Pareto fronts of high dielectric constant plus high bandgap materials.
More from Research
- Language Models Can Rewrite Their Own Weights: 27B Model Self-Improves Without Gradients — Jiajun Fan · 2026-11-21
- Posterior Matching Distillation preserves teacher diversity in few-step generation — ArashVahdat · 2026-11-21
- Lean compiles, but the natural-language proof is still wrong — jasondeanlee · 2026-11-21
- If the Riemann Hypothesis is proved, daily life changes nothing — but 17M lines of Lean say otherwise — stingning · 2026-11-21
- Benchmarking 11 open models' post-GQA attention: up to 139x less memory at 128K — bariskasikci · 2026-11-21
- Stanford talk: Responsible AI in healthcare, from RWE to agentic systems — jonc101x · 2026-11-21