Title: Learning function from structure in neuromorphic networks.
Session: Whole-brain, Connectome-based Models of Brain Dynamics: From Principles to Applications
Speaker: Laura Suárez
Abstract: The connection patterns of neural circuits in the brain form a complex network. Signaling within the network manifests as patterned neural activity and is thought to support the computations that underlie cognition and adaptive behavior. How does the organization of the brain confer computational capacity? One way to address this question is through models that map structural connectivity to emergent neural dynamics and functional connectivity patterns, or to individual differences in behaviour. An alternative way is to directly consider function as a computational property. In this regard, artificial intelligence algorithms offer new ways to study structure-function relationships by conceptualizing function as a computational property. We propose a novel way to address the link between structure and function, but with a focus on computation. To do so, we combine MRI connectomics and reservoir computing to investigate the link between macroscale connectivity, and the computational properties that emerge from network dynamics in the human connectome. Specifically, we construct artificial neural networks endowed with biologically realistic connection patterns derived from diffusion MRI and train them to perform a cognitive task. We provide a use-case driven example that implements a simple memory task to illustrate the power of this approach, and we put forward ideas on how this framework can be extended to study other aspects of the link between structure and function in brain networks.

свернуть