UC San Diego

This dissertation explores the computational role of sleep on memory consolidation. Chapter 1 explores a biophysical model of the thalamo-cortical network as it learns a relational memory task. We conclude by making predictions about the role of sleep on relational memory. Chapter 2 explores a less realistic biophysical model as it learns a digit recognition task. We created a sleep algorithm for this model and show that sleep can improve performance. Chapters 3 and 4 utilize this novel sleep algorithm for biophysical networks and apply this algorithm to artificial neural networks, who suffer from poor generalization (Chapter 3) and catastrophic forgetting (Chapter 4). We show that this sleep algorithm can help mitigate these issues and suggest that sleep is instrumental in memory generalization and continual learning.