Risk-utility tradeoff shapes memory strategies for evolving patterns

Abstract

Keeping a memory of evolving stimuli is ubiquitous in biology, an example of which is immune memory for evolving pathogens. However, learning and memory storage for dynamic patterns still pose challenges in machine learning. Here, we introduce an analytical energy-based framework to address this problem. By accounting for the tradeoff between utility in keeping a high-affinity memory and the risk in forgetting some of the diverse stimuli, we show that a moderate tolerance for risk enables a repertoire to robustly classify evolving patterns, without much fine-tuning. Our approach offers a general guideline for learning and memory storage in systems interacting with diverse and evolving stimuli.

Playing it safe: information constrains collective betting strategies

“…This restricted future utility should limit the value and precision required of the optimal model, while also increasing the imperative for "safety" in light of possible future changes in the world. We can expect trade-offs associated with keeping past memory to play a role in such situations [43,44]. Perhaps this observation also has a bearing on the widespread observation that Occam's Razor, understood as a preference for simplicity or parsimony, seems to be an organizing principle for mental functions [45][46][47][48][49][50], and that biases towards simple models are helpful in many statistical settings that require inference and prediction [22,41,[51][52][53].…”