Learning, prediction accuracy, and neural plasticity in sensory cortex

Abstract

Causal inference during association learning is a cardinal feature of complex nervous systems. In reinforcement learning, a stimulus or context becomes linked to a negative or positive outcome to inform future behavior. Although prefrontal cortex and striatal circuits have been implicated in reinforcement learning, sensory cortex also undergoes marked short-term and long-lasting changes. Here we review studies demonstrating anatomical, synaptic, and task-dependent response plasticity in sensory cortex during learning. A contrast between plasticity induced by sensory association learning, where stimuli predict reinforcement outcomes, and pseudotraining, where sensory inputs are uncoupled, is consistent with sensory cortex's role in prediction evaluation and reinforcement signaling. We propose that plasticity in sensory cortex–a site for collision of internally-generated expectations and incoming sensory input–reflects the relative accuracy of expected versus actual sensory signals as they develop during learning. Sensory learning may thus be a useful tool to probe the function of neocortical circuits.