Published in Computational and Systems Neuroscience (CoSyNe), Mar 2022.
Perceptual learning has been associated with altered sensory cortical representations in trained animals relative to naive ones. While there is substantial variability across animals in the degree of behavioral learning and the associated changes in neural representations, we lack an account of how experiences during learning may drive these differences. Here we address this question in the context of an auditory learning task, by combining experiments and computational modeling. Mice were progressively trained to classify tones as a single, center frequency or non-center by licking left or right, respectively. In parallel, we used calcium imaging to record from a population of layer 2/3 excitatory neurons in the auditory cortex during learning. Despite similar behavioral performance at the end of training, animals exhibited one of two distinct activity profiles in auditory cortex. Specifically, tuning profiles of excitatory neurons exhibited either a relative enhancement or a suppression of responses at the center frequency. We developed a computational model to explore whether animal-specific choice preferences seen during learning could explain this individual variability in neural tuning. We trained a model neural network using reward-dependent Hebbian learning to perform the task, and examined whether initial choice preferences (rates of licking right and left), and the resulting reward statistics, are related to the learned neural representations. We found that higher rates of reward in trials with non-center frequencies early in learning lead to larger magnitude responses to the center frequency, a relationship which was confirmed in the data. Overall, our results suggest that, through its effects on reward statistics and consequent synaptic plasticity, choice preference during early auditory perceptual learning may play a causal role in producing across-animal variability in learned representations.