This Lunch & Learn Series, exclusively tailored for faculty and graduate students within the EECS department, takes place on the last Friday of every month during the fall and spring semesters. A light lunch will be provided by the department. Please RSVP to Angie Chance by April 23.

Join us and engage with our guest speaker, who will highlight campus resources, new initiatives, and other valuable information to foster research collaborations, networking, and other opportunities among our faculty and students.

How AI and HPC are Reinventing Scientific Discovery

Abstract

The scientific method—grounded in prior research, hypothesis generation, prediction, and experimental validation—is increasingly challenged by the overwhelming pace of new scientific publications. This deluge hampers researchers’ ability to stay current, leading to redundant studies and issues with reproducibility. In this talk, I will present a novel approach that leverages advanced AI techniques and high-performance computing (HPC) to automate and accelerate key aspects of the scientific discovery process.

Building on our previous work, which demonstrated how traditional NLP can structure research literature into knowledge graphs and derive hypotheses using linear algebra on HPC systems, we now explore the use of large language models (LLMs) to significantly enhance this pipeline. Our method encodes vast corpora of scientific literature and simulation data into rich knowledge graphs. We then employ graph analytics, knowledge graph embeddings, and graph neural networks to identify promising research directions and generate high-quality hypotheses. These hypotheses will then be validated through simulation, closing the loop in an automated discovery process.

Beyond accelerating research, this framework facilitates the design of ultra energy-efficient circuits optimized for AI algorithms, advancing both the performance and scalability of HPC and edge AI systems. Our approach hopes to transform scientific inquiry by enabling researchers to keep pace with emerging knowledge while fostering innovation in computational hardware.

Biography

Thomas E. Potok, founder and leader of the Computational Data Analytics Group at the Oak Ridge National Laboratory (ORNL), is a distinguished leader in the fields of Data Science and Artificial Intelligence, recognized for his pioneering contributions to machine learning, deep learning, text analytics, and neuromorphic computing. At ORNL, he established and led the Data and AI Research Section, spearheading transformative programs that have advanced these disciplines.

As a Principal Investigator, Potok has secured over $36 million in research funding from leading research organizations, including the U.S. Department of Energy (DOE), Defense Advanced Research Projects Agency (DARPA), Department of Defense (DOD), Department of Homeland Security (DHS), Lockheed Martin, and Battelle. 

His research excellence has been recognized with five R&D 100 Awards, often referred to as the "Oscars of Invention." These accolades highlight his contributions to innovations such as Piranha, a high-performance text analysis system; DTHSTR, a text recommender system; MENNDL, a framework for deep learning system design; SuperNeuro, a large-scale neuromorphic simulation platform; and MAQ, a machine learning system for adiabatic quantum computing.

Beyond his contributions to national research, Potok co-founded VortexT Analytics, where he successfully commercialized Piranha and DTHSTR in collaboration with Covenant Healthcare, leading to the company's acquisition.

He has published over 150 peer-reviewed publications, 17 issued patents, with two additional patent applications. Potok has made a lasting impact on AI and computational science. His achievements have earned him the prestigious designation of Battelle Distinguished Inventor, underscoring his influence and leadership in advancing artificial intelligence and data-driven innovation.