Learning function from structure in neuromorphic networks
- Bratislav Misic, McGill University
How does the wiring of the brain confer cognitive capacity? Technological, analytic, and theoretical advances present a unique opportunity to reconstruct brain networks in individuals. At the same time, modern artificial intelligence algorithms offer new ways to link structure and function, by conceptualizing function as a computational property. Despite common roots, neuroscience and artificial intelligence have followed diverging paths. Recent efforts, spearheaded by our group, present a unique opportunity for convergence of these two vibrant scientific disciplines. We propose to construct bio-instantiated artificial networks, constrained by the real wiring blueprint of the human brain. Our preliminary data suggests that bio-instantiated or neuromorphic networks may provide the most computationally powerful architecture for designing deep neural networks. This research program will build a novel way to study how mental functions emerge from the links and interactions between brain areas. Moreover, our findings will help to identify potential network substrates of brain and mental disorders, thus addressing a problem of major societal importance. By integrating two vibrant and fast-paced scientific disciplines, the program will create unique educational and training opportunities.