Reading specific memories from human neurons before and after sleep

Area of Science:

• Neuroscience
• Artificial Intelligence
• Cognitive Science

Background:

• Episodic memory, the ability to recall single past events, is crucial for human intelligence.
• Decoding specific memories from human brain activity is a significant scientific challenge.

Purpose of the Study:

• To develop a model capable of decoding specific memories from neuronal activity.
• To investigate how memory representations change across brain regions and sleep-wake cycles.

Main Methods:

• Developed a transformer neural network model trained on intracranial electroencephalography (iEEG) data.
• Utilized cross-channel attention to integrate neural spikes across brain regions and timescales.
• Trained participant-specific models to predict memory retrieval of concepts (e.g., persons, places).

Main Results:

• The model successfully predicted memory retrieval of specific concepts from neuronal activity.
• Medial temporal lobe (MTL) activity was crucial for decoding memories before sleep.
• Frontal cortex (FC) activity became essential for decoding memories after sleep.
• Demonstrated a transformation of memory information across neural populations over wake/sleep cycles.

Conclusions:

• Neural representations of memories are distributed across brain regions and dynamically change with sleep.
• This decoding capability offers potential for brain-computer interfaces to modulate specific memories.
• Suggests a pathway for therapeutic interventions in episodic memory disorders.