Photo/ Artem/ Adobe Stock
As artificial intelligence takes on increasingly complex decisions, UC Berkeley researchers are helping rethink the mathematical foundations that make those systems possible. The National Science Foundation has awarded $20 million to renew the National Science Foundation Artificial Intelligence Institute for Advances in Optimization, or AI4OPT, for five years through 2031. Under the cooperative agreement, researchers from institutions including UC Berkeley, Georgia Tech and the University of Southern California will push the convergence of AI and optimization into new territory, with a focus on more scalable, efficient and trustworthy approaches to decision-making and end-use cases in energy systems, supply chains and manufacturing.
UC Berkeley IEOR Professor Alper Atamturk, a co-principal investigator and Berkeley site director for AI4OPT, will lead the institute’s Optimization for AI research thrust. The renewed research agenda challenges traditional boundaries between the two fields: AI-enabled methods will support new approaches to control-aware optimization and optimization proxies, while new convex and nonconvex optimization methods will be developed to improve the training of AI models and more tightly integrate learned models within optimization solvers. These advances will be developed and tested against real-world challenges in energy, supply chains and manufacturing.
“With a new research focus, we are excited to develop novel optimization models and methods, and leverage modern computer architectures to transform AI and learning, enabling decision pipelines that are scalable, trustworthy, and energy-efficient,” Atamturk said.
The UC Berkeley team includes Atamturk, Duncan Callaway, Ying Cui, Paul Grigas and Dorit Hochbaum, bringing together expertise across optimization, machine learning, energy systems and computational methods. The renewed five-year research program will begin Oct. 1, 2026, extending Berkeley’s role in a national collaboration aimed not only at applying AI to optimization, but at reimagining how optimization can shape the next generation of AI.