Publications
Effects of Firm Performance on CEO Compensation Before and During COVID-19
Research in Economics, 2023
This research analyzed how firm performance influenced CEO compensation structures before and during the COVID-19 pandemic, with a focus on changes in pay ratios and incentive alignment.
Research Positions
MABe25 - Animal Behavior Benchmark
Benchmarking video-based AI models for classifying animal behavior across 16 research labs as part of the MABe25 competition.
My work has focused on:
- Large-scale video understanding — evaluating whether pretrained video transformers (VideoMAE2, VideoPrism) can generalize to specialized behavioral science data across diverse lab settings
- Multi-lab benchmark infrastructure — a dataset spanning 16 labs with varied recording conditions, designed to test model robustness for real-world scientific use
AI for Animal Behavior Monitoring - Project accepted to ADSA 2026 Poster!
Building scalable computer vision systems for analyzing dairy calf behavior in agricultural environments, supporting veterinary researchers studying early indicators of disease and welfare.
My work has focused on:
- Self-supervised vision for livestock monitoring — a pipeline using DINOv3 features and YOLOv11 detection to classify calf posture from farm camera footage with limited labeled data
- Bridging AI and veterinary research — an interactive Jupyter workflow enabling everyone to run experiments and interpret results
- Paper: "Automating pose detection in calves: a computer vision framework for non-invasive behavioral monitoring." — Accepted to ADSA 2026 — thank you to all my advisors!
Conducted as part of the Bowers Undergraduate Research Experience (BURE) with support from a CIDA grant.
Research Interests
- Machine learning and computer vision — deep learning systems for real-world perception tasks
- AI for science — using ML to accelerate research in biology, agriculture, and medicine
- Representation learning — self-supervised and foundation models (DINO, VLMs)
- Data-efficient learning — methods that work well with limited labeled data