Uncertainty measure

[levan92]

Hi @AminParvaneh, thank you for sharing your very interesting work! Just want to pick your brains/check my understanding. Assuming we want to come up with an uncertainty measure/score, aka, measure how uncertain an unlabelled sample is, my intuition is that those candidates picked with a smaller alpha cap are more uncertain than those picked later with a large alpha cap. Therefore, alpha norm is an inverse measure of uncertainty. Intuitively, a smaller alpha norm means that an unlabelled sample require less "push" to destabilise its prediction. What do you think?