Abstract
The intertwined effects of recency and practice on memory and learned behavior have long been studied in psychology and neuroscience. We develop a bottom-up, mechanistic theory addressing the combined effects of recency and practice, beginning with a classical single-neuron model, then applying insights from this model to the network level. Combining error-based and associative learning, we mathematically derive the forgetting curve for a single neuron and its dependence on practiced repetition, showing how highly practiced memories or behaviors can become far more resistant to being overwritten by later learning. At the network level, error-based learning is responsible for initial gains in performance, while associative learning gradually transfers control of the downstream population from one input pathway to the other. We interpret the model neurobiologically by identifying the inputs as cortex and thalamus, and the downstream population as striatum, providing a framework for understanding the neural basis of habit formation and the automatization of behavior through practice.
