I work at the intersection of surrogate modeling and scientific machine learning, building data-driven models for complex systems governed by partial differential equations. My experience spans industrial electromagnetics, where I developed SciML surrogates for Maxwell solvers using PINNs, VPINNs/FastVPINNs, and neural operators such as FNO, WNO, and GINO, with GPU-optimized pipelines that significantly reduced training time for large-scale simulations. Recently, my interests have shifted toward transformer-based architectures for PDE generalization and latent-space optimization, including hybrid operator models and the Latent Reciprocity Network (LRN) for robust, geometry-aware generalization across equations.

My current hobbies include cycling, cooking, sketching, and occasionally experimenting with new ways to visualize mathematical and physical concepts through art and diagrams. I am always interested in discussing new ideas and collaborating with people working at the intersection of mathematics, physics, and machine learning.

Mamta Saini