Image from v1.4 (landscapes dataset) (2/22/24)
During my senior capstone (Adversarial Examples) I had access to a department GPU machine for model training. When I was waiting on collaborators or training the capstone project on a different machine, I used this resource to work on a generative model. After the capstone wrapped, I requested special permission to extend my access and keep working on this project. Although I had not learned deeply about GANs I thought it could be a fun challenge to build one from scratch. The results are by no means photo-real but some of the later results are definitely fun to look at. Here are some of the earliest results, I later grew my datasets and improved the models, to see those results, press below.
Image from v1.2 (kingfisher dataset) (2/21/24)
This was my final project for my Machine Learning class. My group and I went about trying to create a function that could take in a chess board position and return an evaluation of win likelihood for the player without ever looking at depth. This would be used as a component in a chess engine and is something I want to look at more for my chess engine project.
Final Capstone Project for Computer Science Major. Explored and compared different adversarial example attacks on image classifiers for 10-week team project. Built and trained image classifiers, developed attack methods, handled dataset curation. Although involved in many parts of this project, I particularly focused on the surrogate model attacks. For this I curated datasets, designed model architectures, and supervised training cycles. In the future I would love to further explore adversarial attacks or defenses against them.