AI-Fun with ELLIS Seminar | Siu Lun Chau: Credal Two-sample Tests: How to Test Hypothesis under Dataset Uncertainty
| Dates: | 12 March 2025 |
|---|
| Times: | 11:00 - 11:00 |
|---|
| What is it: | Seminar |
|---|
| Organiser: | Faculty of Science and Engineering |
|---|
| Who is it for: | University staff, External researchers, Alumni, Current University students |
|---|
| Speaker: | Siu Lun Chau |
|---|
|
March’s AI-Fun with ELLIS Seminar is with Siu Lun (Alan) Chau from CISPA Helmholtz Center for Information Security.
In-person attendance is recommended but for anyone who cannot be physically present, there is a zoom link included below.
Title:
Credal Two-sample Tests: How to Test Hypothesis under Dataset Uncertainty
Abstract:
Dataset-level uncertainty arises when there is ambiguity about the true distribution of interest. This can occur in scenarios involving distribution shifts, when datasets act as proxies, or when multiple datasets are collected under varying conditions. In this talk, we introduce credal two-sample testing, a novel hypothesis testing framework that compares credal sets—convex sets of probability measures that capture both aleatoric and epistemic uncertainty. Unlike classical two-sample tests, which compare precise distributions, our approach supports reasoning about equality, inclusion, intersection, and mutual exclusivity between credal sets. This generalisation allows hypothesis testing to account for epistemic uncertainty directly, addressing challenges posed by partial ignorance stemming from dataset-level ambiguity. We formalise these tests as two-sample problems with nuisance parameters, and we propose the first permutation-based solution for this class of problems. We also present practical, kernel-based implementations, enabling the application of credal two-sample testing to real-world data analysis tasks.
Profile:
Siu Lun (Alan) Chau is an incoming Assistant Professor in Statistical Machine Learning at Nanyang Technological University, Singapore. His research focuses on uncertainty-aware machine learning, explainable machine learning, and preference-integrated learning. Currently, he is a Postdoctoral Researcher at the Rational Intelligence Lab at CISPA Helmholtz Center for Information Security in Germany, where he works with Dr. Krikamol Muandet on imprecise-probabilistic machine learning. Siu Lun completed his DPhil in Statistics under the supervision of Prof. Dino Sejdinovic at the University of Oxford, where his research explored the integration of kernel methods and Gaussian processes, with a focus on their applications to trustworthy machine learning.
We hope to see you there!
Best wishes,
AI-Fundamentals
ELLIS Manchester
Speaker
Siu Lun Chau
Role: Incoming Assistant Professor
Organisation: College of Computing and Data Science, Nanyang Technological University, Singapore
Travel and Contact Information
Find event
3A.057
Nancy Rothwell Building
Booth St East
Manchester
