Projects

• TDAGo: Analyzing AlphaGo Matches using Persistence Homology Python program to analyze Go games using Persistence Homology for my Topological Data Analysis class (Duke Math 713). I used the evolution of persistence diagrams to detect topologically-significant features of games played between iterations of Google Deepmind’s AlphaGo to do rudimentary victor predictions. (TDAGo Code)

• Kaa: Non-linear Reachability Analysis using Dynamic Parallelotope Bundles Software created to experiment with reachable set computations of non-linear systems governed under discrete polynomial dynamics. The project was specifically created to understand the effectiveness of dynamically-reorienting parallelotope bundles on improving the quality over-approximations of reachable sets. It is the first experimental software created to properly plot the evolution of these dynamic bundle strategies for practitioners to understand the efficacy of different bundle strategies. It significantly improved the usability of previously existing reachable set simulators using these techniques. (Dedicated Website Under Construction)

• RNNSPDNet: Riemannian Deep Learning on fMRI images Final Project for Computer Vision (UNC COMP 776). We experimented with Geometric Deep Learning on SPD (Symmetric Positive Definite) matrices arising from BOLD signals derived from Functional MR images. In particular, we formulated an RNN variant of SPDNet to see if it yielded better classification results. Joint work with Juan Garcia. (Code)