Machine Learning for Science
Machine learning (ML) has emerged as a transformative paradigm across numerous scientific disciplines, offering powerful tools to navigate and extract insights from increasingly vast and complex datasets. In fields such as astrophysics, materials science, and engineering, traditional analytical methods often encounter limitations when confronted with high dimensionality, inherent noise, sparsity, or the sheer volume of data generated by modern observatories and experimental facilities. ML techniques, including deep learning, generative models, and probabilistic frameworks, provide solutions for complex tasks such as automated classification, precise parameter inference, anomaly detection, data reconstruction, and the discovery of hidden patterns, thereby accelerating the pace of scientific discovery and enabling new avenues of research.
A significant focus of this research area lies in leveraging ML to tackle grand challenges in cosmology and astrophysics, such as understanding galaxy evolution, mapping the large-scale structure of the universe, and probing the nature of dark energy. These endeavors require robust and scalable methods to process petabytes of observational data, identify rare phenomena like gravitational lenses, accurately estimate astrophysical parameters, and improve the quality of observational data through advanced signal processing. Furthermore, developing interpretable ML models that can provide physically meaningful explanations for their predictions is crucial for building trust and facilitating scientific understanding.
My work is dedicated to developing and applying cutting-edge machine learning methodologies to address pressing challenges in science, with a particular emphasis on astrophysics and engineering. I have developed robust deep learning pipelines for automated detection and modeling of galaxy-scale strong gravitational lenses and have explored galaxy morphology beyond the traditional Hubble sequence using unsupervised machine learning. Furthermore, I have applied generative adversarial networks (GANs) for anomaly detection in Hyper Suprime-Cam galaxy images and other astronomical datasets, proving their efficacy in identifying unusual phenomena. My research also contributes to precise cosmological measurements through the application of deep neural networks for peculiar velocity estimation from the Kinetic SZ effect, reducing model error in weak lensing cluster mass estimation using optimized galaxy selection, and developing SYTH-Z for machine learning synthetic spectra for probabilistic redshift estimation.
Beyond specific applications, I have focused on advancing ML methodologies themselves. This includes enhancing interpretability in generative modeling by developing statistically disentangled latent spaces guided by generative factors in scientific datasets, thereby making complex models more transparent. I have also pioneered the use of neural network-based point spread function deconvolution for astronomical applications, improving image quality, and developed a novel Voronoi tessellation-assisted deep learning approach for global field reconstruction from sparse sensors. My contributions also span engineering applications, including multi-task modeling for engineering applications with sparse data and the application of probabilistic modeling and automated machine learning frameworks for high-dimensional stress field analysis. These efforts provide robust, interpretable, and scalable tools vital for current and future scientific endeavors, including the opportunities presented by projects like the Rubin LSST Dark Energy Science Collaboration.
