Working memory is our ability “keep in mind” information and manipulate it when it is not present in the environment. It is central to many aspects of cognition, such as reading, arithmetic, or simple activities such as keeping track of ideas during a conversation. While several studies have been able to identify the brain networks associated with working memory, the mechanisms through which independent memorized items are maintained and manipulated are unknown.
One hypothesized mechanism is that the brain can reactivate past experiences, like recordings on a tape. To do so, the brain would generate a specific pattern of activity for each memorized item and the replay of this pattern of activity at any subsequent point in time would bring back the associated memory. Replay in the human brain has been reported when people are resting or sleeping. Whether such replay mechanisms are also involved during working memory and their relationship with memory performance are unclear.
In this project, we will apply machine learning techniques to neurophysiological data, to detect working memory brain replays in real-time. Healthy participants will first learn to associate sounds and letters. They will then perform a working memory task based on this association. We will detect when the neural activity patterns related to memorized items are reactivated during the memory task and extract the activity patterns associated with this reactivation. We will then develop a novel brain-computer interface that will be able to detect brain replays in real-time, during task performance, and trigger a non-invasive brain stimulations device. The objective of the brain stimulation is to perform online optimization of reactivation-related brain activity patterns to causally enhance memory performance. This project will thus investigate for the first time the potential causal link between human brain reactivation and memory performance.