Kevin Du

As part of my research project, "Generalizing Backpropagation for Gradient-Based Interpretability", I implemented a semiring-based generalization of machine learning's workhorse: backpropagation. I built this from an open-source, PyTorch-inspired pedagogical library.

I designed and executed a set of baseline experiments to assess the feasibility of using the eliciting latent knowledge technique to extract and erase concepts in large language models without a labeled concept dataset.

I led and mentored a team of 6 students in developing the website and technical tools for Brown University's flagship hackathon in 2019. The 2019 iteration of the website is no longer maintained, but get a sense of the hackathon here!

I led and mentored a team of 6 students in writing and presenting technical workshops throughout the year. Check out an example workshop we wrote and presented to 50+ students on Flask here!