YQI Talk - Sachin Vaidya - MIT

 

AI is playing an increasingly important role in scientific research, with recent successes across a wide range of domains. However, AI-driven scientific discovery faces two fundamental challenges: improving interpretability so that models yield physical insight rather than opaque predictions, and automating experiments to accelerate progress in the lab. In this talk, I will discuss our efforts to address both challenges. First, I will introduce Kolmogorov–Arnold Networks (KANs), a class of neural architectures designed to improve interpretability by combining learnable activation functions with symbolic regression. Unlike conventional black-box models, KANs are transparent and enable direct inspection of learned relationships. I will present applications of KANs in condensed matter physics and photonics, demonstrating how this architecture aligns with various scientific objectives. In the second part, I will focus on a general-purpose automation platform for optical experiments, a longstanding bottleneck in both research and industry. Tabletop optical setups are central to fields ranging from photonics and materials science to biomedical imaging and semiconductor technology, yet they remain highly dependent on manual assembly and alignment. I will present our AI-driven robotic platform, which integrates generative AI, a robotic arm, and computer vision to autonomously assemble and align optical systems. By addressing both interpretability and experimental automation, we take steps towards a future in which AI is an integral and transparent tool in physics for advancing both theory and experiment.
 

References:

[1] KAN: Kolmogorov-Arnold Networks (ICLR 2025)

[2] AI-Driven Robotics for Optics, arXiv:2505.17985 (2025)

[3] Symbolic Learning of Topological Bands in Photonic Crystals, arXiv:2505.10485 (2025)
 

Bio: Dr. Sachin Vaidya is currently a postdoctoral associate at the Massachusetts Institute of Technology (MIT), working with Prof. Marin Soljačić. He is also a junior investigator at the NSF AI Institute for Artificial Intelligence and Fundamental Interactions (IAIFI). He earned his PhD in physics from the Pennsylvania State University in 2023, where he worked on topological photonics. His current research interests include nanophotonics, topological quantum matter, AI for science, and robotics.