An Analysis of Models for Rodent Decision-Making

Abstract

In perceptual decision-making tasks, animals, especially rodents, make blatant errors despite preponderant sensory evidence. This phenomenon, known as 'lapsing', has particularly puzzled scientists, and accordingly numerous theories have been proposed to explain this deviant behavior. Standard models propose that animals employ a single strategy throughout decision-making experiments, using Gaussian noise to simulate inattention. However, recent literature puts forth alternative approaches involving conscious strategy-switching by the rodent. One such approach employs a reinforcement-learning based model to explain lapses as an exploration strategy. Another approach involves characterizing an engaged state to represent correct decisions, and a series of disengaged states to explain lapses. This thesis first replicates these approaches on both simulated data and a rat visual and multisensory experiment data. Then, we propose a new model integrating lapse history with Q-values to fit psychophysical data. The goal is to provide a more holistic explanation of lapses in rodent decision- making.