Research

Machine Learning

Recent advances in computational power and algorithmic innovations have generated a variety of powerful Neural Network architectures; revolutionizing multiple scientific and industrial fields. I am interested in applying these new techniques for improved analysis and discovery in spectroscopy, chemical image analysis, and microscopic imaging. In my recent work, I have improved the spatial resolution of the chemical maps in spectroscopic imaging using SFG. In this hyper-spectral imaging technique, the signal to noise ratio is typically very low; consequently multiple pixels are binned to achieve the required S/N ratio for chemical identification. I have utilized Fully Connected neural networks to extract chemical identity from a single pixel spectrum in extremely low S/N ratio by converting the chemical identification problem into a classification problem. The network is trained on noisy spectra with known chemical labels. After training it can predict the chemical natures of single pixel spectra from unseen samples with more than 92% accuracy. The work is currently submitted to the Optics Communications.

Stochastic Modeling of Interactions in Excitons

Mathematical modeling of stochastic processes particularly thought Ito calculus has played a fundamental role in stock market finance. A similar mathematical problem is encountered in the interaction of excitations and molecules inside chemical and physical systems. These interactions are often stochastic in nature and they influence the line-shape of the spectral signature for the system of interest. Through stochastic modeling, important insights can be obtained into the spectral features generated by these processes. My current research is focused on using numerical and analytical methods to model the stochastic interaction between excitons, calculate their spectral response, and develop machine learning methods to identify stochastic processes in experimental outputs.

My recent presentation at Exciton/Photon Interactions for Quantum Systems 2021 can be found here.

Surface Science & Spectroscopy

Surfaces often show unique physical and chemical properties which are different than that of the bulk. The study, understanding and control of these properties is of fundamental importance in both academia and industry. However, surface typically contains only a fraction of molecules than that in the bulk; consequently, the surface specific signal is overwhelmed by the bulk contribution in most scientific techniques. Furthermore, buried surfaces such as solid-solid interfaces are hard to reach. I am interested in using novel techniques such as Sum Frequency Generation (SFG) spectroscopy, to study the surface properties and their molecular origins for multiple systems of interest including Self-Assembled Monolayers, Organic Semiconductors, chemically terminated silicon surfaces, etc. Our recent work includes the study of molecular arrangement and orientation on Rubrene single-crystal surface using Reflection High-Energy Electron Diffraction (RHEED) and SFG. This molecular characterization is fundamental in understanding the interaction between organic semiconductor - electrolytes for cutting edge devices such as flexible LCDs. Preprint can be provided upon request.

Super-resolution Imaging with Chemical Sensitivity using SFG

The variation of chemical and physical properties across the surface can have profound effects on catalysis, oxidation, wetting, and many other important processes. Chemically sensitive SFG microscopy has proven to be an extremely useful tool for imaging surface chemistry. However, the spatial resolution practically achievable in different SFG microscopies is still in micrometer length scales due to diffraction limit of the mid-IR lasers often used in these systems. I am interested in achieving super-resolution SFG imaging. At Baldelli Surface lab (University of Houston), I have combined the Ground State Depletion (GSD) principle with structured pump-IR; in this scheme, a donut shaped pump-IR is utilized to deplete the SFG relevant vibrational ground state of the chemical surface. The dark center of this optical vortex leaves the ground state population intact in an area that is smaller than the focal spot size. This spot is then probed with a pair of visible and mid-IR beams to generate an SFG signal. Therefore, the SFG signal is generated from a relatively sharper spot than that dictated by the Abbe diffraction limit. The proof of concept experiments using surface hydrogen on silicon show a 3-fold resolution improvement. The preprint can be provided upon request.