CAREER: Untangling Inter-Area Communication in the Brain Using Multi-Region Neural Networks

Human and animal behaviors like learning, remembering, and deciding require the interactions of neurons and circuits across regions of the brain. However, despite the importance of these interactions, remarkably little is known about the processes regulating these brain-wide communications. This research builds computer models of the brain based on measurements taken from humans and animals performing behaviors, and uses those models to identify how different brain regions communicate and work together to produce behaviors. This work will identify shared and distinct features of brain-wide communication to guide new experimental studies and enable new computer models to better define brain functions. Additionally, this project promotes community engagement, diversity, and inclusion through two complementary programs: "Comp-ic Book Neuroscience," which brings research findings from computational neuroscience into under-served classrooms in New York City through the jargon-free and visually appealing medium of comics; and the Student Outreach for Neuroscience Integrated with CS (SONiC) program, an annual lab-based summer school to give NYC-area senior college and graduate students hands-on experience with visualizing and modeling brain data. While rapid advances in neuroscience have catalyzed a deeper understanding of individual brain regions and their functions, these regions generally do not operate in isolation. Yet, little is known about processes regulating the brain-wide communication underlying many behavioral outputs. To reveal fundamental principles of brain-wide communication, this project will produce (1) a new, scalable, robust, and flexible class of multi-region recurrent-neural network (RNN) models with inter-area communication; and (2) analysis methods to infer the direction and magnitude of interactions within and between areas. Multi-region RNNs will be constrained with real neural data to uncover mechanisms of the real biological system, for instance, how the cooperative activity of neurons within and across brain regions gives rise to complex behaviors like decision-making. Reverse-engineering these models will reveal how multi-area brain circuits use biological plasticity to acquire a new skill. Finally, RNN modeling of human electrophysiology data will help identify inter-area communication processes that are conserved or divergent across multiple species. Wider adoption of the new models and tools will transform the understanding of how interacting brain areas function cohesively to orchestrate complex behaviors and inform future experimental paradigms. The research will also promote cross-fertilization between neuroscience and artificial intelligence/machine learning communities, and provide quantitative techniques shared in the broader neuroscience community. Furthermore, the project will foster an inclusive, welcoming environment for a diverse new generation of computational neuroscientists. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.