I build robust visual systems by bridging geometric and probabilistic reasoning with deep learning.
San Francisco Bay Area
We introduce a probabilistic radiance-field framework that preserves 3DGS-level rendering speed while replacing heuristic Gaussian densification with gradient-based optimization of a learnable volumetric density (CVPR 2026).
We introduce a lightweight geometry-aware attention network that learns unrasterized local boundary structure, including curves, corners, and junctions, and generalizes from simple synthetic shapes to noisy low-light photographs (ECCV workshop on Traditional Computer Vision, 2024).
Ph.D., Electrical Engineering — Advised by Prof. Todd Zickler
Thesis: Learning-Based Methods for Recovering Visual Structure
B.S., Electrical Engineering — Data Science Specialization