My focus over the spring and summer was on interpreting active learning queries and some initial attempts at understanding whether we could leverage this interpretability to make active learning algorithms more fair. Talk titled "Interpretable Active Learning" (FAT* via Youtube)"
I'm interested in unsupervised representation learning, model interpretability, and algorithmic fairness. i.e. how can we make the absurd amount of passively generated data work for us without reproducing our biases or causing harm.
Papers and Publications
Throughout undergrad, I've done work in high throughput chemistry using density functional theory calculations and cheminformatic models to speed up the discovery of new organic battery materials. Check out this very cool work I've been a part of as a member of the Schrier lab at Haverford College: "Bio-Inspired Electroactive Organic Molecules for Aqueous Redox Flow Batteries. 1. Thiophenoquinones".
I presented the initial results of my work attempting to understand active learning queries in Sydney at the Workshop on Human Interpretibility in Machine Learning (WHI) "Interpretable Active Learning" (WHI Program).
Contributed Talks and Academic Projects
I recently presented some of my work in fairness, active learning, and interpretability at the "Conference on Fairness, Accountability, and Transparency (FAT*)"
In my sophomore adaptive robotics class, I worked on a project to optimize molecular geometries better than industry-standard software for drug discovery descriptor calculations. "Generating Near-Optimized Molecular Geometries Across Reactions using Neural Networks & Back Propagation" (cs.swarthmore.edu).
I was one of four selected speakers at the Undergraduate Science Symposium, which invites students from ten regional colleges to talk and present posters at Haverford College each year. Delivered talk: "Active Learning Recommendations for Exploratory Synthesis"