Learning hierarchical sequence representations across human cortex and hippocampus.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Humans perceive continuous sensory input as discrete units like words and events.
• Statistical learning is the brain's mechanism for discovering regularities in sequential data.
• Sequence structure can be represented at multiple levels, including transitional probabilities, unit order, and unit identity.

Purpose of the Study:

• To investigate how the human cortex and hippocampus encode auditory and visual sequences.
• To explore the neural tracking of temporal regularities in sensory sequences.
• To understand the hierarchical organization of sequence learning in the brain.

Main Methods:

• Intracranial electrode recordings from human subjects.
• Exposure to auditory and visual sequences with embedded temporal regularities.
• Analysis of neural activity to identify tracking of sequence features.

Main Results:

• Neural tracking of sequence regularities emerged within minutes.
• Early sensory processing areas tracked lower-level features and learned units.
• Later processing stages and associative regions, including the hippocampus, encoded more complex information like unit identity and ordinal position.
• A gradient of representational complexity was observed across brain areas, from transitional probability in early regions to identity in higher-order areas.

Conclusions:

• The brain employs multiple, parallel computational systems for sequence learning.
• These systems are organized hierarchically within cortico-hippocampal circuits.
• Learning rapidly shapes neural representations of sequential information.