My projects sit at the intersection of AI for Science, cybersecurity, scientific machine learning, explainable AI, topological data analysis, high-dimensional visual analytics, and scalable data systems.
Landscaper
A Python framework for exploring the loss landscapes of deep learning models, designed to support diagnostics for scientific machine learning and model interpretability.
Visualizing Loss Functions as Topological Landscape Profiles
Introduces a topological representation for visualizing higher-dimensional loss landscapes, revealing how landscape shape relates to model performance and learning dynamics in tasks such as image segmentation and physics-informed neural networks.
Evaluating Loss Landscapes from a Topology Perspective
Uses topological data analysis to quantify loss landscapes and extract reproducible insights about neural networks, including relationships among topology, performance metrics, Hessian-based measures, and scientific ML behavior.
LossLens: Diagnostics for Machine Learning Models through Loss Landscape Visual Analytics
A visual analytics framework for exploring loss landscapes at multiple scales, integrating global and local metrics to help diagnose architectures and physics-informed neural networks.
ONNX-MLIR
A compiler infrastructure project that transforms valid Open Neural Network Exchange graphs into executable code with minimal runtime support, built on LLVM/MLIR compiler technology.
A Comparison of Decision Forest Inference Platforms from a Database Perspective
A systems study comparing decision forest inference platforms and evaluating when in-database inference can improve end-to-end performance for models such as RandomForest, XGBoost, and LightGBM.
Serving Deep Learning Models with Deduplication from Relational Databases
Introduces storage optimization techniques for serving deep learning models from relational databases, including duplication detection, page packing, and caching to reduce storage cost and inference latency.
Interactive Visualization Server for Geospatial Data Exploration
Part of UCR-STAR, this project supports interactive exploration of geospatial datasets through cache-aware visualization, backend dataset onboarding, and metadata management in MongoDB.