cv

C.H.E.N. Lab (BAIR, CPH)

Undergraduate Researcher

Designed zero-shot LLM pseudo-labeling pipeline to improve semi-supervised learning accuracy. Worked on RadQA dataset; implemented FixMatch on a non-inference task for baseline comparison.

• Self-Supervised Learning
• Machine Learning
• Large Language Models
• Zero-Shot Inference

JLK Inc.

Research Intern

Developed Federated Learning models reaching near identical performance to commercially deployed U-Net models using Python. Collaborated with four M.D. professionals to investigate the use of Federated Learning in medicine.

• Federated Learning
• Machine Learning
• Medical Imaging

Impact AI

Data Engineering Intern

Developed a data preprocessing pipeline to pattern-match raw datasets of various formats from multiple companies using Python. Researched and presented eight AI-based B2B SaaS business case studies, showcasing their strengths, weaknesses, and outlooks.

• Data Preprocessing
• Commercial Budgeting
• Business Analytics

Studio.geo @ UC Berkeley

Undergraduate Researcher

Experimented Progressive-GAN on the Savio cluster to generate artificial maps using Python.

• Generative Adversarial Network
• Machine Learning
• Map Generation

Undergraduate

UC Berkeley, College of Engineering

Electrical Engineering and Computer Science

• CS170 (A+)
• CS188 (A)
• CS70 (A+)
• CS61A/B/C (A+/A+/A)
• EECS16A/B (A+/A+)
• MATH53 (A+)

Random Gradient Masking as a Defensive Measure to Deep Leakage in Federated Learning

Arxiv, Preprint

Compared the efficacy of randomly masking gradients from Federated Learning submissions against other defenses against Deep Leakage from Gradients such as Pruning, Compression, and Noising on Convolution Neural Networks

In-Silo Federated Learning vs. Centralized Learning for Segmenting Acute and Chronic Ischemic Brain Lesions

Medrxiv, Preprint

Comparative analysis of Federated and Centralized Learning on brain MRI images. Showed Federated Learning is as effective as Centralized learning on real-life non-i.i.d. brain lesion datasets of ~10,000 patients over 9 institutions.