ICIP 2026 plenary: generative models reshape priors in computational imaging
prof_kamilov · x · 2026-11-18
Ulugbek Kamilov (UW–Madison) gave an ICIP 2026 plenary, Rethinking Posterior Interface in Computational Imaging — now online (54 min, free for IEEE SPS members).
Core thesis: computational imaging is traditionally framed as ill-posed inverse problems with explicit priors; generative modeling is driving a shift toward learned restoration operators that implicitly encode data distributions — priors are queried via learned models rather than written down. Three key ideas: (1) general restoration operators implicitly define priors; (2) stochasticity is a principled tool for optimization in highly ill-posed problems; (3) priors can be learned directly from measurements without clean training data. The talk unifies proximal methods, plug-and-play algorithms, and modern generative models, with biomedical reconstruction applications.
More from Research
- Ropedia Xperience 2: egocentric video plus metric 3D reconstructions hits HF trending for Physical AI — liuziwei7 · 2026-11-19
- Four slope samples, weights 1-2-2-1: a visual explainer of why Runge-Kutta 4 is so accurate — burny_tech · 2026-11-19
- I8-Project Uses AI Agents to Crack Long-Open Higher-Curvature Calculations in 11D Supergravity — burny_tech · 2026-11-19
- Robot RL discussion: reset buffer design is an overlooked unsolved problem — YouJiacheng · 2026-11-19
- ICLR authors get zero reviewer responses to rebuttal with scores 3,3,2 — pastaHacker · 2026-11-19
- IIT Delhi professor wins Google grant for multi-agent safety and interpretable defense — Tanmoy_Chak · 2026-11-19