Our paper "Towards Sustainable Scientific Machine Learning Fast and Interpretable PDE Solvers via RBF-PIELM" has been accepted at ACM CODS 2025 as oral and archival work !

Excited to share that I’ll be presenting my research, “Towards Sustainable Scientific Machine Learning: Fast and Interpretable PDE Solvers via RBF-PIELM” at the ACM India Joint International Conference on Data Science (hashtag#cods 2025) !

In the domain of AI for Science, we need PDE solvers that are not just accurate, but also fast and sustainable. Our work introduces Radial Basis Function-based Physics-Informed Extreme Learning Machines (RBF-PIELM), a lightweight, mesh-free alternative to traditional Physics-Informed Neural Networks (PINNs).

Key Contributions:

• 𝗠𝗮𝘀𝘀𝗶𝘃𝗲 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆 𝗚𝗮𝗶𝗻𝘀: RBF-PIELM achieves orders-of-magnitude reduction in runtime and energy usage by replacing PINNs’ time-consuming gradient descent with a single-shot least-squares solve.
• 𝗦𝘂𝘀𝘁𝗮𝗶𝗻𝗮𝗯𝗶𝗹𝗶𝘁𝘆: Using the Green Algorithms Methodology, RBF-PIELM requires significantly less energy and produces substantially lower carbon emissions compared to PINNs. This is a major step towards sustainable scientific computing.
• 𝗦𝘂𝗽𝗲𝗿𝗶𝗼𝗿 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲: For comparable solution accuracy, RBF-PIELM trains up to 350x faster and requires up to 13x fewer parameters than PINNs.
• 𝗜𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆: The approach offers greater interpretability through physics-aware kernel initialization.

This research lowers the hardware barrier for AI in science, enabling broader accessibility for researchers with limited compute resources.

If you are at CODS 2025, Let us chat about the future of hashtag#AI4Science, hashtag#SustainableAI, and hashtag#PhysicsInformedML!

I’d like to extend my sincere gratitude to my mentors, Dr. Vikas Dwivedi, and Prof. Balaji Srinivasan, for their guidance on this research.