For the Greater Good of All

Abstract

Stress and genetic background regulate different aspects of behaviorallearningthrough the action of stress hormones and neuromodulators. In reinforcementlearning (RL) models, meta-parameters such as learning rate, futurereward discountfactor, and exploitation-exploration factor, control learning dynamicsandperformance. They are hypothesized to be related to neuromodulatorylevels inthe brain. We found that many aspects of animal learning and performancecan bedescribed by simple RL models using dynamic control of the meta-parameters.Tostudy the effects of stress and genotype, we carried out 5-hole-boxlight conditioningand Morris water maze experiments with C57BL/6 and DBA/2 mouse strains.The animals were exposed to different kinds of stress to evaluateits effects onimmediate performance as well as on long-term memory. Then, we usedRL modelsto simulate their behavior. For each experimental session, we estimateda setof model meta-parameters that produced the best fit between the modeland theanimal performance. The dynamics of several estimated meta-parameterswerequalitatively similar for the two simulated experiments, and withstatistically significantdifferences between different genetic strains and stress conditions.