Generative replay underlies compositional inference in the hippocampal-prefrontal circuit

Area of Science:

• Cognitive Neuroscience
• Neurobiology

Background:

• Human reasoning relies on combining information in novel ways.
• The neural mechanisms of compositional computation remain largely unknown.
• Understanding how the brain implements complex reasoning is a key challenge.

Purpose of the Study:

• To investigate the neural basis of compositional computation in the human brain.
• To explore how the brain constructs novel representations from constituent elements.
• To examine the role of neural replay in hypothesis testing during reasoning.

Main Methods:

• Functional magnetic resonance imaging (fMRI) to identify brain regions involved in object construction.
• Magnetoencephalography (MEG) to analyze neural replay dynamics during problem-solving.
• Task-based experiments requiring participants to assemble wholes from given elements.

Main Results:

• fMRI revealed relational and compositional representations in the frontal cortex and hippocampus for novel objects.
• MEG demonstrated that neural replay sequences assemble elements into compounds, acting as hypotheses.
• The content of replay sequences dynamically evolved, progressing towards the correct solution.

Conclusions:

• Neural replay in the hippocampal-prefrontal circuitry supports compositional inference and hypothesis testing.
• These findings suggest a computational bridge between memory and reasoning functions.
• Generative replay plays a crucial role in constructing complex representations and solving problems.