Research
My research focuses on developing and applying computational methods at the intersection of astrophysics, cosmology, and machine learning. The work spans foundation models for scientific applications, advanced ML techniques for astronomical data analysis, cosmic structure investigation, and statistical inference methods.
Disclaimer: This section is automatically updated by Reasoning Language Models. Google Gemini is utilized to periodically go over my recent publications, talks and activities to update the content. While the information is monitored, at times incorrect information may appear.
Foundation Models for Science
Foundation models are transforming scientific research, enabling specialized AI assistants. Methods include developing multi-modal foundation models for cosmological data analysis and domain-adapted large language models (LLMs) for astronomy and spectroscopy. These specialized LLMs achieve benchmark-topping performance, often matching or exceeding general models in scientific Q&A and reasoning tasks. Key is fine-tuning with domain-specific data to interpret scientific language and complex datasets, facilitating intelligent assistance for cosmological ensembles and deeper analysis. Establishing rigorous evaluation methodologies is crucial for assessing their utility as scientific research assistants, demonstrating profound impact across diverse disciplines.
Learn more about this research →
Learn more about this research →

From: Multi-modal Foundation Model for Cosmological Simulation Data
7 Publications 1 Figures Available
Machine Learning Applications in Science
Machine learning is transforming scientific research by addressing complex data challenges, particularly in astronomy and cosmology. Key methodologies include deep neural networks for tasks like astronomical image deconvolution, galaxy morphology exploration using unsupervised learning, and robust parameter estimation for redshifts and peculiar velocities. Generative Adversarial Networks (GANs) are frequently applied for anomaly detection in observational data and for benchmarking AI-evolved cosmological simulations. Researchers also leverage probabilistic modeling for high-dimensional stress fields and sparse data, develop differentiable models for large-scale structure predictions, and enhance interpretability in generative models. These applications collectively accelerate discovery, improve data analysis from massive datasets like Rubin LSST, and enable deeper understanding of complex physical phenomena.
Learn more about this research →
Learn more about this research →

From: Application of probabilistic modeling and automated machine learning framework for high-dimensional stress field

From: Benchmarking AI-evolved cosmological structure formation

From: A Modular Deep Learning Pipeline for Galaxy-Scale Strong Gravitational Lens Detection and Modeling
18 Publications 12 Figures Available
Dark Matter, Cosmology & Large-Scale Structure
Recent research in dark matter, cosmology, and large-scale structure employs diverse methods to unravel cosmic mysteries. Studies utilize cosmic shear analyses to constrain modified gravity theories and leverage multi-stream phase-space views to detail the dark matter web's intricate caustic structures and halo substructure. Cosmological simulations, including auxiliary-variable-guided generative models and galaxy formation frameworks like CRK-HACC, model the interplay between dark matter and baryons. Observational efforts span photometric mapping of stellar populations in the Milky Way, to large-scale surveys of galaxy clusters via the SZ effect (SPTpol), and future spectroscopic missions like SPHEREx. These approaches enhance our understanding of structure formation, dark matter properties, and fundamental cosmological parameters.
Learn more about this research →
Learn more about this research →

From: Multi-stream portrait of the Cosmic web

From: Topology and geometry of the dark matter web: a multistream view

From: The Caustic Design of the Dark Matter Web
11 Publications 5 Figures Available
Emulation, Inference & Uncertainty Quantification
Emulation, inference, and uncertainty quantification (UQ) research advances scientific discovery by employing sophisticated machine learning. Key methods include Gaussian Process (GP) emulators and Probabilistic Neural Networks (PNNs), often integrated into Reduced-Order Models (ROMs). These surrogates efficiently map complex input parameters to system outputs, accelerating inference and enabling exploration of vast parameter spaces across domains. Applications span cosmology (e.g., subgrid models, modified gravity) and fluid dynamics. A critical focus is robust, interpretable UQ; PNNs inherently quantify predictive uncertainty, while dedicated approaches ensure interpretability in AI for high-energy physics. This work delivers efficient, reliable, and interpretable predictions vital for complex scientific modeling.
Learn more about this research →
Learn more about this research →

From: Matter Power Spectrum Emulator for f(R) Modified Gravity Cosmologies

From: Emulator-Based Inference of Cosmological Subgrid Models

From: Probabilistic neural networks for fluid flow surrogate modeling and data recovery
6 Publications 5 Figures Available
