The paper delves into the concept of realisability in statistical learning theory under the assumption of epistemic uncertainty. The author considers a scenario where the train and test distribution are derived from the same credal set, a convex set of probability distributions. The study is presented as a first step towards a more comprehensive treatment of statistical learning under epistemic uncertainty. It also touches upon the concepts of domain adaptation and generalisation in machine learning.
Publication date: 22 Feb 2024
Project Page: https://arxiv.org/abs/2402.14759v1
Paper: https://arxiv.org/pdf/2402.14759
