PosteriorBench Accepted at NeurIPS, Redefining Evaluation for Inverse Problems
PosteriorBench, accepted at NeurIPS 2026, argues that solvers of scientific inverse problems should be evaluated on recovering full posterior distributions rather than single-point reconstructions, since high reconstruction accuracy does not guarantee accurate uncertainty estimation.
2026-09-24 ~ 2026-09-25 · 2 related posts
- PosteriorBench: better reconstruction accuracy can mean worse posterior recovery — AnimaAnandkumar · 2026-09-24
- PosteriorBench accepted to NeurIPS: point reconstruction accuracy misleads on inverse solvers — AnimaAnandkumar · 2026-09-25