Research & Innovation
My research develops AI and machine learning methods for high-stakes scientific problems — with a particular focus on rare-event prediction in space weather and AI-driven acceleration of physics-based simulations. This work spans the full pipeline from raw scientific data to deployable, open-source ML tools used by the broader research community.
Core Research Areas
AI & Deep Learning for Predictive Systems — Neural architectures (CNNs, GANs, GNNs, Transformers) for solar flare and SEP prediction; explainable AI (XAI) methods including Grad-CAM and attention attribution for operational forecasting; synthetic data generation to mitigate extreme class imbalance
Neural Operator Surrogate Modeling — SFNO-based surrogates for accelerating MHD simulations across heliospheric and plasma physics domains, reducing simulation cost by orders of magnitude
ML from Imbalanced & Rare-Event Datasets — Robust sampling strategies, novel evaluation metrics (Contingency Space), loss function design, and shapelet discovery for extremely skewed scientific datasets
Multivariate Time Series Analytics — MVTS preprocessing toolkits, temporal similarity metrics (TS-MIoU), time series clustering, outlier detection, and Wavelet/Fourier decomposition methods
Spatiotemporal Data Mining — Discovery of event sequences and co-occurrence patterns from trajectory data; localized outlier detection; spatiotemporal interpolation and data quality assessment (SMART framework)
Scientific Cyberinfrastructure & Open Science — Design of large-scale solar data testbeds and repositories integrating image and time-series data; open-source ML toolkits (MVTS-Data Toolkit); AI-ready benchmark datasets (SWAN-SF) serving the global space weather community
Information Retrieval & Data Indexing — Content-based image retrieval (CBIR) for large solar image archives; high-dimensional indexing (iDistance, KNN); semantic data integration via ontology-guided techniques; dimensionality reduction for efficient search and classification
Research Deliverables
- Publications — 200+ peer-reviewed works spanning IEEE T-PAMI, Scientific Data, Astrophysical Journal, and more
- Grants & Contracts — $18M+ in secured R&D funding from NASA, NSF, and industry across 20+ awards
- Open Software — Tools including the MVTS-Data Toolkit for multivariate time series preprocessing
- Open Data — Public datasets including the SWAN-SF Dataset for space weather ML benchmarking