​​​Talk from Yunzhe Liu - Centre for Neuroimaging; Max Planck UCL Centre for Computational Psychiatry

Short abstract:

​​​Our brains can flexibly combine existing information to create new knowledge, a process known as compositional computation. While this fundamental operation underlies wide-ranging human cognition and flexible behaviour, its neural mechanisms remain poorly understood. It is thought to involve the hippocampal-prefrontal circuit. To investigate this, we conducted intracranial recordings in both the hippocampus and cortical areas simultaneously in the human brain. Subjects performed two LEGO-like tasks, constructing silhouettes using a set of building blocks. We found that hippocampal ripples coordinate medial prefrontal cortex (mPFC) activity to compositionally represent the silhouettes as relational configurations of their building blocks. When inferring the silhouette construction, hippocampal ripples were linked with neural replays that sampled task-dependent sequences of their constituent elements (i.e., building blocks). The strength of these replays during the ripple period predicted subsequent inferential performance, with this relationship mediated by mPFC activity during the same ripple period. These findings suggest that hippocampal ripples act as a computational bridge between the hippocampus and mPFC, enabling the inference of new knowledge by compositionally recombining existing elements.​