Md Nasim

Postdoctoral Associate, Cornell University

Department of Computer Science,

Ithaca, NY, 14853, USA.

Phone: 765-609-1997

I am currently a Postdoctoral Associate in Cornell University, Ithaca, NY (Department of Computer Science). I am working with Professor Carla Gomes on research related to AI for Climate Institute and Cornell University AI for Science Institute (CUAISci) in support of the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship program, a program of Schmidt Futures.

Before joining Cornell, I got my PhD degree in 2024 from Purdue University, West Lafayette, IN. During my PhD research, I worked on designing efficient methods for accelerating scientific discovery of new knowledge from data. I developed a collection of methods to address the lack of precise modeling, lack of efficient learning methods and lack of human-in-the-loop integrated frameworks in AI-driven scientific discovery workflow. My work has been deployed in the real world for rapid analysis of experimental videos and has led to the scientific discovery of a new property of material defects under extreme heat and irradiation.

Prior to joining Purdue, I worked as a lecturer in the Department of Computer Science and Engineering in United International University, Bangladesh. I received a BSc degree in 2015 from Bangladesh University of Engineering and Technology, Bangladesh.

LinkedIn profile.

Research Highlights

Physics Knowledge Discovery via Neural Differential Equation Embedding

To address the lack of precise modeling of physics learning tasks, we developed Neuradiff, an end-to-end method for learning phase field physics models from noisy video data. In previous learning approaches involving multiple disjoint steps, errors in one step can propagate to another, thus affecting the accuracy of the learned physics models. By encoding the physics model equations directly into learning, end-to-end Neuradiff framework can provide ~100% accurate tracking of material defects and yield correct physics model parameters.

Efficient Learning of Sparse and Decomposable PDEs using Random Projection

To address the lack of efficient methods for partial differential equation (PDE) physics model learning, we developed Rapid-PDE method. We scale up learning these first-principle models harnessing randomized algorithm, exploiting the fact that the temporal evolution of many physical systems happens at the "small" interface of the system components and these changes are sparse. By exploiting this sparsity, Rapid-PDE can reduce PDE model learning times by 50-70%.

Unraveling the size fluctuation and shrinkage of nanovoids during in situ radiation of Cu by automatic pattern recognition and phase field simulation

To address the lack of human-in-the-loop integrated frameworks for scientific discovery, we developed a framework to rapidly analyze terabytes of in situ irradiation experiment videos. In situ experiments are essential for analyzing material behaviors under extreme heat and irradiation, and designing sustainable materials. Using our framework, we were able to accelerate experimental video data analysis. Detailed analysis enabled by our framework led to the scientific discovery of the surprising size fluctuation of nano-size void defects in irradiated materials.