Leading the effort on developing a graph-based generative adversarial network, which we call MPGAN, which has proven effective at generating sparse data with irregular underlying geometry. Our latest work on this has been accepted to the 2021 NeurIPS main conference. We're now experimenting with a conditional GAN version and variable-sized graphs, as well as applications to other datasets such as CERN detector data.
Developing a library and website for convenient access to jet datasets, in particle cloud representations, along with several useful utilities for jet-based machine learning development. Has been used so far for our group's MPGAN and LGAE projects, and continues to be expanded.
Wrote a review of deep learning models that are equivariant to physics-relevant group transformations for Prof. John McGreevy's fantastic group theory course. This led to our group developing a graph-based autoencoder equivariant to Lorentz group transformations (LGAE). Preprint on our latest work is in progress.
Developing graph-based autoencoders for compression of and anomaly detection in particle cloud representations of Large Hadron Collider data. Our latest work has been accepted to the Machine Learning for Physical Sciences workshop at NeurIPS 2021.
Created dynamic, sub-micron holographic optical tweezers and a Quantum Gas Microscope with sub-micron resolution in order to manipulate individual atoms (or qubits) for quantum computing and quantum information science experiments. This work won a William A. Lee Research award, and will be published soon.
Fun project mostly to gain experience with RNNs and Attention. I achieved a 71% testing accuracy in predicting the outcome of European soccer matches.
Created an app for visualizing course requirements with a user-friendly UI. I was the Back-end and Algorithms Lead for a team of 10, and personally wrote the server, scraping and graphing algorithms for the app. We were one of 8 finalists out of 60 projects in the UCSD 2018 software engineering course.