NEURAL PROCESSING OF HIGH-ORDER AUDITORY STRUCTURES IN HUMANS

In nature, successive sensory inputs are correlated, their relations ranging from local transitions to complex hierarchical structures. Humans learn these statistical regularities efficiently, even from limited or incomplete exposure. Previous work has shown that humans effortlessly learn complex modular graph structures and that their performance aligns with an associative learning model that incorporates memory decay. However, the neural coding supporting such rapid learning and the mechanistic implementation of the hypothesized model are still underdefined. To investigate this, we recorded both whole-brain intracranial EEG and single unit recordings in the hippocampus in humans listening to auditory sequences generated by such modular graph structures. Preliminary results revealed a fast response to sound position in the graph, around 150ms, compatible with a surprise response elicited by associative learning mechanisms. This response was present in hippocampal regions, but also in the rest of the temporal lobe and in prefrontal regions. Multivariate analysis will further allow us to study the neural geometry of the sequence representation at different learning stages. At term, our work will map both the macro- and microscopic neural signatures of network structure learning in humans.