Distinct roles of hippocampus and neocortex in symbolic compositional generalization

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Humans possess a remarkable ability to combine symbols, generating novel meanings and enabling complex cognition.
• Understanding the neural underpinnings of this compositional ability is crucial for deciphering higher-level cognitive functions.

Purpose of the Study:

• To investigate the regional neural mechanisms supporting the combination of discrete symbolic features for spatial inference.
• To explore how the brain represents and combines visual attributes (shape, color) to form novel representations.

Main Methods:

• Participants performed a spatial inference task involving the combination of shape and color features.
• Blood-oxygen-level-dependent (BOLD) functional magnetic resonance imaging (fMRI) was employed to measure brain activity.
• Multivariate pattern analysis (MVPA) was used to decode neural representations of stimuli.

Main Results:

• The hippocampus demonstrated high-dimensional, parallel encoding of visual attributes, facilitating flexible individuation.
• Neural patterns for composite stimuli in the ventromedial prefrontal cortex (vmPFC), posterior parietal cortex (PPC), and primary visual cortex (V1) were predictable from element patterns.
• Composition in vmPFC occurred in a high-dimensional format, whereas PPC and V1 utilized a low-dimensional, spatially aligned frame of reference.

Conclusions:

• The findings elucidate the distinct neural strategies employed by different brain regions in supporting compositional generalization.
• This research provides novel insights into the neural circuits enabling the flexible combination of information for complex reasoning.