Project Summary Memory reactivation, which occurs during both wakefulness and sleep, is a leading candidate mechanism for how new information is consolidated to form long-term memories. Influential theories propose complementary roles for how reactivation benefits consolidation during both non-rapid eye movement (NREM) and rapid eye movement (REM) sleep?NREM sleep strengthens new memories in cortical networks (systems consolidation) and REM sleep reorganizes memories within local networks (synaptic plasticity). Over the last two decades, myriad studies have contributed to our understanding of NREM sleep and systems consolidation. The role of REM sleep, however, largely remains a mystery, possibly because prior work has not been tuned to detect REM sleep-dependent memory effects. We propose that REM sleep plays a crucial role in promoting representational change, specifically by decreasing neural overlap between related memories (differentiation), which reduces competition during subsequent memory retrievals. We will conduct experiments, complemented by neural network modeling, to test how competitive learning and memory reactivation during NREM and REM sleep affect how memories relate to each other in representational space and how much they compete during subsequent learning and retrieval. Our first aim is to run experiments where we experimentally manipulate levels of competition during wakeful learning. To track evidence of learning-specific memory reactivation during subsequent sleep, we will ?inject? decodable information into the learning experience and use machine learning classifiers to distinguish these learning conditions from sleeping EEG brain activity alone. We will use fMRI to track competitive dynamics and quantify representational change, and demonstrate the behavioral consequences of this differentiation?less neural overlap should reduce interference and promote new learning. Our second aim will use computational models to test predictions about how neural representations change during NREM and REM sleep, and how these changes impact behavior. The model will be tuned using Aim 1 experimental data, and will generate predictions to inform Aim 3 experiments. Our third aim is to run experiments where we experimentally bias the learning-related content of reactivation during NREM sleep. We will examine how our NREM reactivation manipulation modulates reactivation during subsequent REM sleep, and whether REM sleep is necessary for this NREM manipulation to have an effect on memory outcomes. The proposed studies use a novel combination of state-of-the-art approaches, including: (1) machine learning to decode memory content from sleeping EEG brain activity; (2) multivariate pattern analysis of fMRI data to track representational change; (3) behavioral measures sensitive to representational change; (4) neural network modeling to identify the mechanisms underlying NREM- and REM-related plasticity; and (5) NREM-based targeted memory reactivation to test how memory processing during NREM and REM sleep interact.