Abstract representations emerge in human hippocampal neurons during inference.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Human cognition relies on abstract representations for generalization.
• Neural encoding of these abstract representations, especially during learning, remains poorly understood.
• Understanding how the brain forms and utilizes abstract representations is key to explaining complex cognitive functions.

Purpose of the Study:

• To investigate the neural encoding of abstract representations during inferential reasoning.
• To characterize the representational geometry in various brain regions, including the hippocampus.
• To determine how learning influences the formation of these neural representations and their relation to behavior.

Main Methods:

• Recorded neural activity (single units) from neurosurgical patients performing an inferential reasoning task.
• Analyzed representational geometry in the hippocampus, amygdala, medial frontal cortex, and ventral temporal cortex.
• Examined how learning through trial-and-error or verbal instruction impacts neural representations.

Main Results:

• Hippocampal neural representations uniquely encoded multiple task variables in an abstract, disentangled format.
• This disentangled representational geometry emerged specifically after learning to perform inference.
• Both learning methods (trial-and-error and verbal instruction) resulted in similar hippocampal representational geometries.
• A strong correlation was observed between the abstract representational format and inference behavior.

Conclusions:

• Abstract and disentangled representational geometries in the hippocampus are critical for complex cognitive tasks like inferential reasoning.
• Learning plays a pivotal role in shaping these neural representations.
• The findings provide insights into the neural basis of generalization and adaptation in humans.