Representations of common event structure in medial temporal lobe and frontoparietal cortex support efficient...

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Modeling

Background:

• The brain utilizes cognitive maps for memory and inference, recognizing common structures across contexts for adaptive behavior.
• Understanding how abstract representations of common structure are formed and used for novel decisions is crucial.

Purpose of the Study:

• To investigate the neurocomputational mechanisms underlying the extraction and utilization of abstract representations of common structure.
• To examine how the brain forms and uses these representations to support novel decision-making.

Main Methods:

• A neurocomputational approach was employed with human participants.
• Participants learned image pairs (AB, BC) from distinct triads (ABC) with shared internal structure.
• Behavioral data (response times) and neuroimaging data were analyzed to infer indirect associations (AC).

Main Results:

• Hippocampal and frontoparietal regions formed abstract representations with common geometric structure across triads, even without explicit reinforcement.
• Parahippocampal cortex representations were hierarchical, distinguishing between and integrating cross-triad relationships.
• Computational modeling revealed dissociable vector-based retrieval and pattern-completion processes, supported by distinct brain regions (hippocampus for pattern completion; parahippocampal and parietal cortex for vector retrieval).

Conclusions:

• Common geometric structure in neural representations provides a vector space for consistent coding of inferred relationships, facilitating faster inference.
• Distinct neural processes, pattern completion and vector-based retrieval, mediate inference and are supported by specific brain networks.
• This study elucidates how the brain generalizes knowledge across contexts through abstract, geometrically structured representations.