MCCHE Precision Convergence Webinar Series with Raghu Machiraju
Moving towards Interpretable Models of Functioning Brains For Adaptive Real-World Behavior and Sustainable Health
Raghu Machiraju
Professor of Biomedical Informatics, Computer Science and Engineering (CSE), and Pathology at the Ohio State University (OSU)
With High-Level Panel of Leaders in Science, Technology, On-the-Ground Action, and Policy
As exemplified by the Virtual Brain Platform that is part of the EU-lead e-Brain initiative, the brain can be best modeled as a complex, dynamic and adaptive system that is adept at engaging in non-trivial decision making or learning tasks. In previous work we have shown that appropriate state- space models can depict how the various regions of a functioning brain are recruited in a cascacadic manner to complete mental arithmetic. These models were used to explain the different functioning of normal, dyscalculic and dyslexic brains. I will describe this work in sufficient detail and especially highlight the crucial role of state-space models. In more recent interactions, I explored how these state-space models can be used for more general decision making tasks. The goal was to create “models similar to a functioning brain” rather than replicate the brain as is often done in the annals of neuromorphic computing. When successful, it will be then possible to simulate the behavior of humans when they interact with both natural and managed-natural systems. Thus, the models of experts can be shared to teach novices particular tasks. Further, I will also propose that is possible to model living systems of human brain and body as integrated into a larger environment as they adapt to change. Thus, cancers like glioblastomas, living plant(s), and other organisms can be modeled as adaptive precision care. Interestingly, one could create specialized virtual E-brains and realize them either as von-Neuman systems at the least, or as material systems (e,g.,photonic systems emulating photo-synthesis) that sense, compute and store information and interact with each other. I therefore offer a blueprint for an e- brain-in-a-box which, in turn, will rest on many tools from difference projects assembled in high-performing computing integrative architecture like C-Brain.
About the speaker
Raghu Machiraju is a Professor of Biomedical Informatics, Computer Science and Engineering (CSE), and Pathology at the Ohio State University (OSU). He founded the $170M, 55-faculty strong, Translational Data Analytics Institute dedicated to the adoption of data science and analytics on the campus of Ohio State. Currently, he is the Associate Chair for Growth in the Department of Computer Science and Engineering and an essential member of a leadership team overseeing tremendous growth in size and reputation. Over the last two years, CSE@OSU has risen 11 spots and is now a top-25 department and is seeking to transform itself into a school of computing. Raghu’s own research interests span areas where computing intersects with various domains. As a Co-PI of a $20M NSF-funded AI Institute, he helps with the adoption of AI by various domain specialists while contributing to AI foundations. As an independent researcher, he has contributed to developing machine learning methods to characterize unsteady flow, model state transitions of a functioning brain, integrate multiple omics data to predict patient outcomes with both semin- supervised and unsupervised tools, create weakly supervised models that rely on weak labels and enable robust grading of large whole slide histopathology images, and develop tools of GenAI to convert text describing branching processes to flow graphs.
About the series
The Precision Convergence series is launched to catalyze unique synergy between, on the one hand, novel partnerships across sciences, sectors and jurisdictions around targeted domains of real-world solutions, and on the other hand, a next generation convergence of AI with advanced research computing and other data and digital architectures such as PSC’s Bridges-2, and supporting data sharing frameworks such as HuBMAP, informing in a real time as possible the design, deployment and monitoring of solutions for adaptive real-world behaviour and context.
The Precision Convergence Webinar Series is co-hosted by The McGill Centre for the Convergence of Health and Economics (MCCHE) at McGill University and The Pittsburgh Supercomputing Center, a joint computational research centre between Carnegie Mellon University and the University of Pittsburgh.