About me

Hi! I’m Duncan Soiffer, a 2024 - 2025 (December) Master’s in Machine Learning student at Carnegie Mellon University. Previously, I was an undergraduate at Worcester Polytechnic Institute from 2020-2024, where I majored in computer science and math. I am broadly interested in all aspects of machine learning, and especially interested in core methods.

I recently worked in Professor Virginia Smith’s lab, where I led a project on extending agreement-based cascading to generative language tasks (where agreement is difficult to quantify). This resulted in a paper submitted to EMNLP. Currently, I am working in Professor Pradeep Ravikumar’s lab on causal representation learning. On the side, I’m also experimenting with exploring inductive bias and the solution space of representation learning methods through a novel (but as-of-yet unproven) lens, and a project designing search algorithms for a constructive mathematical proof.

Besides this, I have previously worked on reinforcement learning, AI for science, agentic AI, and a cubesatellite. Out of all of my projects, I’m most proud of my Major Qualifying Project (undergraduate thesis) at WPI. Along with fellow undergraduates Andrew Salls and Chase Miller, and supervised by Professors Daniel Reichman, Gábor Sárközy, and George Heineman, we improved the best known lower bounds on the Chvátal–Sankoff constants (click here for our paper) and additionally created a website to help teach and explore the Longest Common Subsequence problem more generally.

(Fun fact: I have an Erdős number of 2!)