We present a novel approach for test-time adaptation via online self-training, consisting of two components.

Evidential Deep Learning (EDL), grounded in Evidence Theory and Subjective Logic (SL), provides a robust framework to estimate uncertainty for out-of-distribution (OOD) detection alongside traditional classification probabilities.

When such components are expected to be embedded in safety-critical systems (e.g.,

Providing guarantees on the decision-making processes of autonomous systems, often based on complex black-box machine learning models, is paramount for their safe deployment. This need motivates efforts towards responsible artificial intelligence, which broadly entails questions of reliability, robustness, fairness, and interpretability.

This highlights an urgent need of contributions from the machine learning community.