(2fo) Dynamic Catalysts: Machine Learning Assisted Operando Characterization.

My interdisciplinary research interest in materials science and chemical engineering is focused on developing next generation of nano-engineered optoelectronic devices and dynamic catalysts using 1) time-resolved operando studies for next-generation of mixed scale 0D-2D hybrid catalysts, 2) correlating structural and excitonic dynamics, and 3) machine-learning models based “inversion” of experimental spectra to structural and electronic descriptors.

Postdoctoral Research:

In my postdoctoral research with Prof. Anatoly Frenkel in Structure and Dynamic of Nanomaterials Lab at Materials Science and Chemical engineering I worked in two major developments in X-ray absorption spectroscopy a) development of machine learning assisted methods for analyzing X-ray absorption spectroscopy and b) application of time-resolved operando XAS studies to study surface restructuring dynamics of Pd-Au bimetallic nanocatalysts under reaction conditions.

Design and characterization of “dynamic catalysts”, enabled by rational catalyst design utilizing a combined theoretical-computational-experimental approach, is at the forefront of heterogenous catalysis for various industrially relevant catalytic processes. The dynamic nature of the “active” sites in these nanocatalysts can be tuned under operando conditions thus enabling the design of new surface ensembles which can go beyond the classical barrier of catalyst performance. The challenges in achieving the precise control over surface ensembles requires a daunting task of understanding the changes in atomic environment of active species due to stimulus and reaction conditions, which can vary time and length scale. I developed an approach which enables periodic stimulus-based characterization of nanoparticle to identify surface structural dynamics of dilute metal component acting as the catalytic centers.

Inverting XAFS spectra to structural descriptors and mapping the evolution of changes in these descriptors during the in-situ experiment or under varying experimental conditions are of significant interest to catalysis community. Previous work has shown that “inverting” XAFS spectra to structural descriptors is possible using artificial neural networks. Machine learning (ML) enabled XAFS analysis has provided new tools to analyze the structural and electronic structure information from X-ray absorption spectra. Currently, the most used ML based “inversion” approach is supervised in nature and it also require development of system specific training databases using theory driven tools. My work has shown that low dimensional representations of data make it easier to extract hidden patterns and these representations can then be further used to build classifiers and other predictors. These low dimensional latent representations can also be utilized to invert the experimentally obtained XAFS spectra to structural and electronic properties of the catalysts e.g. Pd nanoparticles under hydrogen atmosphere at elevated temperatures. In my postdoctoral work, I applied Autoencoders for developing these low-dimensional representation in an unsupervised manner and utilized Variational Autoencoders (VAE) and its variants which can help disentangle non-linear interactions between underlying explanatory factors.

Doctoral Research:

During my Ph.D. research, I pursued research on structure-property characterization of energy harvesting materials using advanced optical characterizations tools and developed a novel, low-cost micro-patterning techniques for conjugated polymers. My PhD research work was conducted at the Center for Functional Nanomaterials (CFN) at Brookhaven National Lab (BNL) focused on investigating structural, optical and electronic properties of semiconducting nanocrystals, 2D materials and their hybrids using ultrafast spectroscopy methods. Specifically, I examined the ultrafast dynamics (pico-seconds to nano-seconds) of photoinduced interfacial interactions between organic and inorganic nanomaterials with sensitivity down to single particle level.

Teaching Interests

I have previously designed and taught an undergraduate research course that introduces freshmend undergraduates to scientific methodology and introduction to research and engineering. My tenets for teaching strategy, while honed during my teaching experience as a Lecturer in Discipline of Physics teaching “Frontiers of Science” course for Columbia College freshmen utilizing active learning classroom strategies and collaboration with Center of Teaching and Learning at Columbia University, are also borrowed from my prior science communication initiatives. During my training on teaching and mentoring in form of a graduate school course and an online course of “Scientists teaching Science” offered by NYAS, I learned that students have different styles of learning. I would be very interested in developing course focused on advanced data analysis techniques using machine learning in materials science and chemical engineering. I would also develop a course on advance materials characterization methods utilizing synchrotron facilities and ultrafast spectroscopy methods.