Biography

Anima has made fundamental contributions to AI that is revolutionizing scientific modeling and discovery. She invented Neural Operators for learning multiscale phenomena that frequently occur in nature, such as fluid dynamics, material modeling and wave propagation. She employed Neural Operators to train the first AI-based high-resolution weather model, tens of thousands of times faster than existing physics-based forecasting. Her AI algorithms have enabled many other scientific advances such as modeling plasma evolution in nuclear fusion, enabling safer autonomous drone flights, and designing novel medical devices, drugs, and functional enzymes. Earlier in her career, Anima spearheaded the development of tensor methods, probabilistic latent variable models, and analysis of non-convex optimization. Anima is Bren Professor at Caltech. She previously was a Senior Director of AI Research at NVIDIA and Principal Scientist at Amazon Web Services. She received her B.Tech from IIT Madras, and her Ph.D. from Cornell University. She did her postdoctoral research at MIT and an assistant professorship at UC Irvine. She has received several honors such as the IEEE fellowship, Alfred. P. Sloan Fellowship, NSF Career Award, and Faculty Fellowships from Microsoft, Google, Facebook, and Adobe. She is part of the World Economic Forum's Expert Network.

SRP Project Title

Neural Operators for Scalable and Sustainable Scientific Modeling

NAIRR Project

Aligning AI models for scientific simulations under a physics-informed framework

Topical Areas

Applied Computer Science; Applied Mathematics; Artificial Intelligence and Intelligent Systems; Computer Science; Other Computer and Information Sciences

Abstract

Addressing global sustainability challenges requires fast, accurate, and generalizable modeling of complex systems in energy, climate, and infrastructure. Traditional high-fidelity simulations are often too slow and computationally expensive for timely exploration, design, and decision-making. This project leverages and advances neural operator frameworks to accelerate scientific simulations while adhering to known physical laws. By combining the expressiveness and inference speed of deep learning architectures with physics knowledge, these neural operators provide predictive surrogates for systems governed by partial differential equations, enabling scalable, energy-efficient computation at previously inaccessible scales. In this research, embedding physical constraints ensure sreliable predictions, generalizes modeling across diverse domains, and enables inverse design to identify system configurations that meet performance goals. Uncertainty quantification and formal verification in Lean (a theorem prover) provide further guarantees of correctness and reliability. Physics-informed enhancements, operator-based multi-scale learning, and robust modeling strategies ensure accurate long-term behavior and broad applicability to complex real-world systems. These tools support high-impact simulations for renewable energy, environmental resilience, and critical infrastructure planning. By releasing open-source frameworks, fostering accessibility, and prioritizing usability, this work empowers scientists, engineers, and students worldwide to harness advanced neural operators for transformative discovery, informed decision-making, and sustainable, verifiable solutions to urgent global challenges.

Desired Skills

• Knowledge of machine learning fundamentals
• Strong programming skills and experience with scientific computing and deep learning libraries (e.g., Python, PyTorch)
• Familiarity with differential equations and numerical methods
• Interest in computational modeling and applying AI to scientific problems
• Enthusiastic and proactive in exploring new research questions, methods, and learning opportunities
• Open-minded and adaptable, eager to engage with diverse scientific approaches and perspectives
• Passion for interdisciplinary research bridging AI and physical sciences
• Effective communication skills
• (Optional) Familiarity with theorem provers such as Lean or an interest in learning them can be helpful for certain specific directions, though not required for most directions

Lightning Talk Title

Neural Operators for Scalable and Sustainable Scientific Modeling

Keywords

AI for science; neural operators; physics-informed ML; accelerating simulations and scientific discovery; inverse design;