ECODE Lab

Starting in 2025 and continuing through 2026, I worked as a research assistant in the ECODE Lab, contributing to ongoing projects at the intersection of evolutionary computation and AI interpretability. I first found out about the lab after taking the graduate algorithm design course taught by the principal investigator of the lab. I talked with her and was able to connect with a PhD student who I shared similar interests with. We were both interested in Jeff Clune's work and the promsie of evolutionary computation being a major part of AI development. Although his proejcts weren't directly related, it was enough for me to get interested in the work and join the lab.

Over my time there, I worked on two projects. The first was a neural architecture comparison between evolution strategies vs gradient descent. This project was meant to scale the research finding's of the fractured engtagled hypothesis to larger datasets like MNIST and CIFAR-10. I worked on buiilding tooling to analyze and visualize model internals, running experiments, and presenting them. This led to interesting findings that have been handed off the the PhD student.

The other project was related to fooling AI models. My PhD mentor's work was building a suite of different ways of fooling AI models. He tasked me with creating a new one. I developed a new algorithm for fooling image recognition that focused on being quick and having real world usage. Again, I built my own tools, ran large-scale exerpiemnts, and presented my findings. This work led a paper in collaboration with two professors that is currently being published.