CS&E Alum Anuj Karpatne Featured at White House Event for AI Research
“The focus of my NAIRR pilot award is to build a new class of foundational models for aquatic sciences, termed Lake-GPT, to model a variety of processes related to the quality of water in freshwater bodies by using novel advances in the emerging field of ecology knowledge-guided machine learning (eco-KGML),” said Karpatne. “With support from the National Science Foundation (NSF) , we will be leveraging the supercomputing facilities at Oak Ridge National Laboratory (ORNL) to train AI models that help us analyze the effects of climate change and land use policies on the quality of water in lakes and reservoirs and even project their future health based on our current actions.”
The Lake-GPT NAIRR project is a collaboration between Karpatne and other computer scientists and ecologists from Virginia Tech, University of Wisconsin - Madison, and ORNL, with NEON, EPA, and USGS as partnering institutions. This six-month project has been granted 750,000 Graphical Processing Unit (GPU) hours on the Summit supercomputer at ORNL. Karpatne was among the 10 award recipients invited to present their work at the White House OSTP event announcing the launch of the NAIRR program. Karpatne was also one of the two recipients invited to give a longer talk on their projects at the AI Expo for National Competitiveness hosted by the Special Competitive Studies Project the following day.
“Our work is opening a new chapter of research in AI to model scientific systems where we have growing volumes of data collected from sensors as well as generated from model simulations. Inspired by the success of large-scale foundation models in mainstream applications of computer vision and language understanding in the commercial arena, the domain of aquatic sciences is ripe with opportunities to investigate the effectiveness of foundation models for modeling scientific processes using both scientific knowledge and data.”
Karpatne’s research on building Lake-GPT makes a distinct departure from mainstream practices of building “black-box” AI models, that solely rely on supervision contained in data, to also leverage the wealth of scientific knowledge available in many domains including aquatic sciences (e.g., conservation laws of mass and energy). This is the framework of knowledge-guided machine learning (KGML) that was developed at the University of Minnesota while Karpatne was working as a PhD student with Regents Professor Vipin Kumar.
“I owe everything to my advisor Vipin Kumar and the graduate program at UMN,” said Karpatne. “This entire field of knowledge-guided machine learning emerged while working on the Climate Change Expeditions Project led by Vipin. This is when we got introduced to problems in climate science where there is wealth of knowledge in the form of models and equations, which when coupled with ML models provide novel grounds for improving the generalizability and scientific consistency of ML results. Our early work in this area laid the roadmap for research in the rapidly growing field of KGML that is now being explored in a wide range of disciplines across academia, industry, and national labs. KGML provides novel opportunities to make fundamental advances in AI/ML fueled by the needs of scientific disciplines and to accelerate scientific progress in interdisciplinary problems of high societal relevance. All of my current projects at Virginia Tech were made possible by my work in graduate school.”
Learn more about the NAIRR pilot program and Karpatne’s work in their feature on Science.org.