Aditi Krishnapriyan, a faculty scientist in Berkeley Lab’s Applied Mathematics and Computational Research Division (AMCR) and assistant professor at UC Berkeley, has been awarded a prestigious 2025 U.S. Department of Energy Early Career Research Program (ECRP) award.
With this support, she will develop innovative, scalable machine learning methods that enable fast and accurate predictions grounded in real-world science, creating a system that balances accuracy, efficiency, and scalability for complex scientific problems. Now in its fifteenth year, the ECRP award supports exceptional researchers during the critical stages of their formative work by funding their research for a period of five years.
At UC Berkeley, Krishnapriyan has faculty appointments in the Department of Chemical and Biomolecular Engineering and the Department of Electrical Engineering and Computer Sciences. She is also affiliated with the Berkeley Artificial Intelligence Research Lab and the Bakar Institute of Digital Materials for the Planet.
As modern scientific research increasingly relies on massive datasets generated by advanced computer simulations and cutting-edge experiments, machine learning offers a powerful way to extract insights. However, current methods often face practical challenges when handling large-scale scientific data because they require more computational power than is readily available in practice and can struggle to deliver accurate, physically consistent predictions.
Krishnapriyan’s ECRP project, Accelerating Large-Scale Atomistic and Continuum Simulations with Physically Consistent and Scalable Machine Learning Methods, aims to overcome these challenges by developing machine learning models and frameworks that can efficiently scale as the dataset size and complexity grow. While it’s a common trend to embed physics constraints directly into machine learning models, her research explores whether training models on large, high-quality data allows them to learn these constraints implicitly a strategy that could lead to even greater scalability. She is also developing methods to speed up predictions while ensuring they remain accurate and consistent with the underlying scientific principles. The project will rigorously validate these approaches across diverse scientific domains characterized by complex spatiotemporal dynamics and significant computational demands, aiming to create a robust framework that balances accuracy, efficiency, and scalability.
“Science currently has access to a lot of simulation data, and machine learning is already helping researchers extract valuable insights and accelerate discoveries," said Krishnapriyan. "But I believe there’s still untapped potential – especially in developing machine learning methods for simulating large, complex systems over long time scales. For example, simulating transport properties in electrolyte materials or running molecular dynamics or fluid dynamics simulations often involves very large systems and extended time frames. These kinds of simulations can currently take months to produce results relevant to real-world experiments. If we can reduce that simulation time to just a week with high accuracy, it would fundamentally change scientific workflows and speed up discovery."
Read the full announcement
- Lawrence Berkeley National Laboratory (Berkeley Lab): AMCR’s Aditi Krishnapriyan Receives 2025 Early Career Award
