Computational modelling of synergistic information in artificial and biological neural networks

How does the human brain orchestrate information processing to enable cognition? Disentangling qualitatively different kinds of information has revealed that the brain regions supporting complex cognition interact synergistically: the information they provide together goes beyond the sum of their individual contributions. Humans also exhibit greater brain synergy than non-human primates, raising the prospect of harnessing synergy to understand human intelligence, and guide the development of AI systems. To realise this potential, we must understand how synergy emerges from the brain's complex organisation, and its computational advantages and drawbacks. Does synergy come at a cost of increased vulnerability to injury or disease? To fill these gaps, I will design AI architectures optimised to exhibit high synergy, and assess their computational capacity and limitations. Leveraging large-scale neuroimaging datasets, I will investigate how brain synergy is reshaped across health and disease, obtaining an information-specific characterisation of human cognition. I will develop an innovative combination of biophysical modelling and 'neuromorphic' neural networks constrained by the brain's empirical wiring: systematically perturbing this joint model with disease-realistic lesions will dissect which features of brain organisation shape the resulting patterns of synergy. Overall, this project aims to establish an information-theoretic lens to illuminate biological and artificial intelligence.