Some of My Projects
We took a deep dive into a SOTA defense against adversarial attacks which uses duality as a proof of defense against adversarial agents. This is a draft.
Cal Hacks ‘23:
Eye tracking and sentiment analysis from your own Zoom conference! Sentieye uses XGaze for eye tracking which we then calibrate on top of your screen to visualize what catches your attention. Code release soon.
AISF ‘23 Hackathon Winner
Clarity enables object segmentation and removal for your videos using META AI’s Segment Anything Model (SAM) and diffusion models. Imagine: adblock for your own personal videos! Code release soon.
A kinda unorganized repo that documents the different experiments and tutorials I complete from time to time. Andrej Karpathy’s a big advocate for documenting your learning as you go, so this is a step in that direction! For some of the research papers I’ve read, you can find their implementations in the
implementations directory. Soon: Obsidian integration for my research paper system.
Scraping for re(search) opportunities.
A python script with BeautifulSoup4 to scrape ML conferences and query for any professor, university, or topic of choice. I had made this to find potential research advisers through their published work rather than their (relatively outdated) personal pages.
A project by Launchpad at Berkeley, Phodexr lets you provide your own image and potential captions, after which our model will find the most likely caption! I was in charge of development of the image embeddings and deplyoment, using over 400k images across COCO Captions and other datasets.
Gait Classification with STMicroelectronics’s STM32Cube
Upon completing UCLA’s ENGR 96C IoT course, several of my classmates and I utilized the SensorTile for gait classification between stair ascent, descent, and neutral walking. A blog post of our work can be found here and the technical report can be found here.