AI/ML
I am interested in the potential of AI to accelerate scientific investigations across disciplines through data-driven insights, predictive modeling, and intelligent automation. My current interests span both the application of AI/ML to real-world scientific problems and the development of novel AI methods that are robust, interpretable, and informed by physical principles.
I am especially excited about multi-modal and physics-informed approaches that integrate diverse sources of information—such as experimental data, simulations, and domain knowledge—into unified learning frameworks. These hybrid methods hold great promise for advancing our understanding of complex systems in areas like materials science, chemical imaging, and quantum dynamics.
Beyond applications, I am equally interested in exploring fundamental challenges in AI, including learning under uncertainty, generalization from limited data, and interpretable decision-making. My long-term goal is to develop AI systems that not only perform well in controlled settings but also translate effectively into real-world scientific and engineering contexts.
Below are some of my works in these directions:
ChemSpecNet: Deep Learning for Hyper-Spectral Chemical Imaging
Optics Communications, Volume 507, 15 March 2022, 127691.
ChemSpecNet is a deep learning framework I developed to bring modern computer vision and machine learning techniques into the field of chemical imaging. It was designed to address a core limitation of Sum Frequency Generation (SFG) spectroscopic imaging: the need for spatial averaging (pixel binning) to overcome low signal-to-noise ratios, which traditionally comes at the cost of spatial resolution.
SFG imaging is a uniquely powerful method for probing surface chemistry, but its weak signals often demand long acquisition times or heavy post-processing. Conventional methods like spectral curve fitting break down in noisy environments and are computationally expensive. ChemSpecNet tackles this challenge by reimagining the problem as a spectral classification task. It uses a supervised neural network to directly identify chemical signatures from noisy pixel-level spectra, enabling high-resolution imaging without compromising detail or speed.
Trained on over a million spectra from Self-Assembled Monolayers (SAMs) on gold substrates, ChemSpecNet achieves:
92% classification accuracy at the single-pixel level (no binning)
Up to 99.5% accuracy using minimal 8×8 binning
Robust generalization across experimental variations
Full-resolution, real-time chemical mapping without the need for long acquisition times
Technically, ChemSpecNet is built as a fully connected neural network for Hyper-Spectral imaging using TensorFlow, with:
Input: mid-IR SFG spectra with 71 wavenumbers per pixel
Outputs: Chemical identities of pixels and chemical maps for full image.
This project shows the power of data-driven models in domains traditionally governed by physics-based approaches. ChemSpecNet opens new possibilities for fast, high-resolution chemical imaging in materials science, nanotechnology, and biomedical sensing—setting a new standard for applying machine learning in hyperspectral and spectroscopic imaging.
Image generated from 1x1 binning using ChemSpecNet
Image generated from 8x8 binning using ChemSpecNet
Surrogate Modeling of Plasma Response in Metal Nano-particles & Inverse Design
In Preparation
Artificial Intelligence and Machine Learning for Chemical Identification
In Preparation