Anima Anandkumar Is Accelerating Scientific Discovery with AI

Summary

Scientific progress is often limited not by a lack of new ideas, but by the cost and complexity of testing them. New solutions are needed to make that testing easier—and researchers like Anima Anandkumar are leading the way. She has conducted cutting-edge research across academia and industry for over a decade, pioneering new AI algorithms that simulate physical systems with unprecedented speed and accuracy—in some cases, over a million times faster than traditional methods. By empowering AI to model these systems, her research has unlocked advances across science and engineering, from high-resolution weather forecasting to designing novel medical devices. “What fascinates me is how to bridge the gap between theory and practice, because I started at a time when deep learning wasn't there—you had to start from first-principles design methods,” says Anandkumar, who explains that her approach to designing algorithms builds on fundamental principles found in maths and physics. Anandkumar is the Bren Professor of computing and mathematical sciences at Caltech, where she leads the Anima AI + Science Lab. She has also worked as a principal scientist at Amazon Web Services, designing machine learning-based solutions for Amazon cloud and a senior director of AI research at Nvidia. Informed by other scientific domains, particularly physics, she says her focus has always been on making algorithms “more principled, hardware efficient, and robust.” Starting from this first-principles approach, Anandkumar and her collaborators developed “neural operators”: a kind of universal AI framework that can learn to simulate physical processes across multiple scales, from molecular interactions to climate patterns. Unlike large language models such as ChatGPT, AI models built with this framework can incorporate the laws of physics to test the plausibility of their predictions. And unlike traditional methods of simulating physical processes, which require immense computational resources to perform millions of calculations from scratch for each new prediction, these models are able to “learn shortcuts” from the data they’re trained on, Anandkumar explains—allowing them to simulate processes with equal or greater accuracy than methods that rely on raw computation, but at a much faster pace. Models designed in this way are particularly powerful because they “have the flexibility to le…