Rui Ponte Costa

Title: Brain-wide credit assignment: cortical and subcortical perspectives

Abstract: The brain assigns credit to trillions of synapses remarkably well. How the brain achieves this feat is one of the great mysteries in neuroscience. Recently, we have introduced Bursting cortico-cortical networks, a computational model of hierarchical credit assignment that captures a large number of biological features while approximating deep learning algorithms (Greedy et al. NeurIPS 2022). I will show that in contrast to previous work this model (i) does not require a multi-phase learning process, (ii) is consistent with experimental observations across multiple levels and (iii) provides efficient credit assignment across the cortical hierarchy.

However, these models often assume that behavioural feedback is readily available. How the brain learns efficiently despite the sparse nature of feedback remains unclear. Recently we have proposed that a subcortical region, the cerebellum, predicts behavioural feedback, thereby unlocking learning in cortical networks from future feedback. We have introduced two views by which the cerebellum may help the cortex: (i) by driving cortical plasticity (Boven et al. Nature Comms 2023) or (ii) by driving cortical dynamics (Pemberton et al. bioRxiv). Together these two views suggest that cortico-cerebellar loops are a critical part of task acquisition, switching, and consolidation in the brain.