Diffusion models go beyond images: now powering driving policies and protein design
jasondeanlee · x · 2026-11-19
At a Simons Institute talk, Grigoris Chrysos (UW-Madison) showed diffusion models are used far beyond image generation:
- Self-driving: generating diffusion policies — from scene encoding plus a Gaussian distribution, denoising yields multi-mode trajectory predictions (citing Liao et al. 2025).
- Protein design: denoising from Gaussian noise to produce de novo antibody structures (citing Bennett et al. 2026).
- His talk, "Grounding Diffusion Models: The Path to Reliable Synthesis Without Hallucination," addresses making diffusion synthesis reliable and hallucination-free.
Diffusion is emerging as a general-purpose policy/structure generation paradigm across domains.
More from Research
- Nimo Robotics launches N1 glove and D1 hand: highest-DoF commercial UMI system for industrial robot skills — Scobleizer · 2026-11-19
- Debiasing Without Annotations: Cross-View Consensus Boosts Worst-Group Accuracy — Weiwei Li · 2026-11-19
- KAIST Proposes PROLA: Low-Rank Adaptation for Spatiotemporal Weather Downscaling — MLVLab · 2026-11-19
- NTU's PACT Adapts Time-Series Foundation Models Without Forgetting Old Conditions — NanyangTechnologicalUniversity · 2026-11-19
- The Resolution Hypothesis: agent evolution stalls because noisy evals can't spot real improvements — ShanghaiTech · 2026-11-19
- MIT's QDA uses LLM agents to auto-design quantization, beating best public recipes — mit-han-lab · 2026-11-19