Unifying Approaches in Active Learning and Active Sampling via Fisher Information and Information-Theoretic Quantities (2208.00549v2)

Abstract: Recently proposed methods in data subset selection, that is active learning and active sampling, use Fisher information, Hessians, similarity matrices based on gradients, and gradient lengths to estimate how informative data is for a model's training. Are these different approaches connected, and if so, how? We revisit the fundamentals of Bayesian optimal experiment design and show that these recently proposed methods can be understood as approximations to information-theoretic quantities: among them, the mutual information between predictions and model parameters, known as expected information gain or BALD in machine learning, and the mutual information between predictions of acquisition candidates and test samples, known as expected predictive information gain. We develop a comprehensive set of approximations using Fisher information and observed information and derive a unified framework that connects seemingly disparate literature. Although Bayesian methods are often seen as separate from non-Bayesian ones, the sometimes fuzzy notion of "informativeness" expressed in various non-Bayesian objectives leads to the same couple of information quantities, which were, in principle, already known by Lindley (1956) and MacKay (1992).