cv

Graduate Student Researcher

CITRIS @ UC Davis Health

• Researching CV methods including object detection, segmentation, and pose estimation in healthcare.
• Utilizing NLP and LLMs for early dementia detection through language-based analysis.
• Developing machine learning algorithms for human activity recognition using video and vibration data.

Teaching Assistant

Department of Electrical and Computer Engineering, UC Davis

EEC 174AY-BY Applied Machine Learning

• Designed lab assignments focused on CNNs, medical segmentation, and object detection applications.
• Delivered guest lectures and coding demonstration on model optimization and pruning techniques.
• Mentoring design teams from project development to implementation, offering hands-on technical assistance.

Undergraduate Research Assistant

RUBINET Research Lab

• Evaluated pruning methods to optimize Low-Light Image Enhancement (LLIE) models for real-time inference.
• Compared magnitude and gradient-based pruning methods, noting efficiency trade-offs.
• Proposed custom loss functions for pruning, reducing model size while retaining quality.

M.S. Computer Engineering

University of California, Davis

B.S. Electrical Engineering

University of California, Davis

Dementia Detection from Audio Signals

• Developing novel framework leveraging LLMs for early dementia detection.

Health Applications Targeted Compression of Vision Models

• Automating video annotation and aligning actions to time-series sensor data.

Optimizing and Compressing Deep Learning Models for Image Preprocessing (LLIE)

• Assessing pruning techniques for image restoration methods.
• Benchmarking LLIE techniques for noise reduction, color restoration, edge deployability (FLOPs).

Benchmarking MLLMs for Visual-Lingual Hallucinations and Bias

YOLOv11, GroundingDino, SAMv2, PyTorch

• Created counterintuitive dataset to evaluate MLLMs’ reliance on language vs. visual input.
• Developed test cases exposing lingual bias in multimodal models without visual context.
• Analyzed MLLM responses to counter common-sense images, identifying hallucination.

Benchmarking Low-Light Image Enhancement (LLIE) and Restoration

PyTorch, LLIE (CNNs, ViTs, Diffusion), YOLOv9

• Benchmarked LLIE models to boost image quality and support downstream vision tasks.
• Implemented YOLO on preprocessed images from LLIE models, validating its performance on low-light dataset.
• Implemented the solution in video footage, highlighting its potential for deployment in security systems.