A new agreement between Los Alamos National Laboratory and the National Science Foundation’s Artificial Intelligence Institute for Advances in Optimization, or AI4OPT, at Georgia Institute of Technology will drive research in applied artificial intelligence and engage students and other professionals in the future of the burgeoning field.
“This collaboration will help develop new artificial intelligence technologies for the next generation of scientific discovery and the design of complex systems and the control of engineered systems,” said Russell Bent, scientist at Los Alamos. “At Los Alamos, we have a lot of interest in doing work related to the optimization of complex systems, and we see an opportunity with AI to make systems more resilient and efficient in the face of factors like climate change, extreme events and third-party actors.”
The agreement spells out a research and educational partnership that will focus especially on AI tools for advancing a next-generation power grid. Building, maintaining and optimizing the energy grid requires extensive computation, with the time and power that that computation entails. AI-informed approaches, including modeling, could efficiently and effectively address power-grid problems.
An AI approach to optimization and problem-solving
Optimization is the practice of finding solutions that can effectively and efficiently use resources. The new research partnership uses the expertise and resources at Georgia Tech to develop “trustworthy foundation models” — models that, incorporating AI, save on the vast computing needed by some models to solve complex problems.
In a system like energy grids, optimization means quickly sorting through possibilities and resources and delivering immediate solutions for power distribution in a crisis. The research agreement will develop “optimization proxies” that build on current optimization methods by using broader parameters, including generator limits, line ratings, generator commitments and grid topologies. While artificial intelligence approaches have helped train optimization proxies for energy applications, doing so using broader parameters has so far been a significant research challenge.
The collaboration will also concentrate on solving problems posed by the wide range of missions and applications at the Laboratory. The team’s research will build upon pioneering efforts in graph-based, physics-informed machine learning to help address Laboratory mission problems.
Outreach and training opportunities
The Laboratory will also host a Grid Science Winter School and Conference in January 2025, with dozens of students and postdoctoral researchers among the participants in multiple days of lectures on methods and techniques applicable to the electrical grid by Lab scientists and its academic partners. This year, with Georgia Tech as a co-organizer and partner, artificial intelligence optimization for the energy grid will be front and center at the event.
The Laboratory began collaborating with Georgia Tech in 2020, with a focus on the energy grid. A number of industrial and academic organizations are part of AI4OPT, including Los Alamos. The institute’s mission is to facilitate “a paradigm shift in automated decision making at massive scales by fusing AI and Mathematical Optimization, to deliver breakthroughs that neither field can achieve independently.”
“The use-inspired research in AI4OPT seeks to address fundamental societal and technological challenges,” said Pascal Van Hentenryck, AI4OPT director. “The energy grid is of special importance as a complex system that is central to our everyday life. Working with Los Alamos, we have an opportunity to advance a research mission and educational vision that makes an impact for science and our society.”
The three-year agreement runs through 2027 and is funded as a Laboratory Directed Research and Development program director’s initiative project on advancing Artificial Intelligence for Mission, or ArtIMis, which in turn supports the Laboratory’s signature institutional commitment in support of AI. Earl Lawrence is the project’s principal investigator, with Diane Oyen and Emily Casleton joining Bent as co-principal investigators.
Bent, Castleton, Lawrence and Oyen are also members of the AI Council at the Laboratory. The AI Council is charged with helping the Lab navigate the fast-evolving landscape around AI, build investment capacities, and forge industry and academic partnerships.
As emphasized in the announcement of the Department of Energy’s Frontiers in Artificial Intelligence for Science, Security and Technology (FASST) initiative, AI technologies will broadly transform the contributions of the laboratories to national missions. This AI4OPT partnership with Georgia Tech builds key strengths for that future.
LA-UR-24-28889