https://fanyi-kaitlyn-wu.github.io/tags/uncerairnty-quantification/Uncerairnty QuantificationHugo Blox Builder (https://hugoblox.com)en-usSun, 01 Feb 2026 00:00:00 +0000https://fanyi-kaitlyn-wu.github.io/media/icon_hu_982c5d63a71b2961.pnghttps://fanyi-kaitlyn-wu.github.io/tags/uncerairnty-quantification/https://fanyi-kaitlyn-wu.github.io/blog/wuml_estonia/Sun, 01 Feb 2026 00:00:00 +0000https://fanyi-kaitlyn-wu.github.io/blog/wuml_estonia/<p>I am happy to attend and present our work in Workshop on 2026 Uncertainty in Machine Learning (WUML).</p> <p>📈Conformal Prediction tells us how often we’re right. ⚛️Bayesian models tell us how uncertain we should be. Instead of treating them as separate tools, our work brings them together.</p> <p>🌟The key idea is simple but powerful: instead of relying on fixed heuristics, we frame conformal prediction as a decision risk problem, using Bayesian integration to optimise efficiency under guaranteed coverage.</p> <p>The goal? Keep prediction sets small and stable, while still guaranteeing coverage.</p> <p>What makes it stand out: • It stays reliable even when the Bayesian model is misspecified • It behaves well under distribution shift (e.g. ImageNet-A) • The result is uncertainty estimates that are not just valid, but usable</p> <p>Many thanks to my collaborator Veronika Lohmanova, and my supervisors Michele Caprio, Samuel Kaski for their guidance and support. If you have any comments, thoughts, and questions, please let us know!</p>