Resource Rational Analysis of Memory Use in Bayesian Updating

Bayesian inference has been used to model many aspects of cognition, and this paper aims to investigate memory using a computational perspective. If memories are costly, how should a learner allocate its resources and still make accurate predictions? We present a framework for an agent that seeks to learn the true weight of a coin after observing a fixed number of observations. The agent pays to retain a noise-minimal belief and increase prediction accuracy. The beliefs of the agent reside in a hierarchical structure, and we hope to see how different types of information should be managed using a resource-rational strategy.