Aidan Lorenz
PhD candidate in mathematics and aspiring AI/ML Engineer/Data Scientist. aidanlorenz@gmail.com
Projects
Generative AI in Robotics
In Progress
This project, joint with Ramy Yammine, aims to utilize generative AI architectures (currently we are using variational autoencoders) to improve upon the current industry standards for sampling-based motion planning in robotics (RRT, RRT*, PRM, etc.). The idea is that these methods rely on randomly sampled points in the configuration space of the robot and we believe that generative AI can help us sample these points more intelligently than simply uniformly at random.
Evaluating the Real-World Safety and Robustness of Deep Learning Models
As part of the Math-to-Industry Bootcamp capstone internship project, my group (Kean Fallon, Jessie Loucks-Tavitas, Sandra Tsiorintsoa, Benjamin Warren, and I) and I worked with Charlie Godfrey and Henry Kvinge from Pacific Northwest National Laboratory to assess the robustness and safety of various deep learning models. The models we studied included Segment Anything Model (an image segmentation model developed by Meta) and GPT-2 (an earlier version of OpenAI’s popular GPT large language models) among others. To see the results of this research-style project, have a look at the white paper linked above.
BrewSavvy Beer Recommender System
As part of the Erdos Institute Data Science Bootcamp group project, my group (Timothy Alland, Brandon Butler, Phuc Nguyen, and I) built a beer recommender system using matrix factorization and variants thereof. See the attached executive summary and Github pages for more details.
Baseball Hall of Fame Voting Predictor
GitHub Repo
Paper (from 2019 - the original version of the project)
As part of our undergraduate senior capstone course in mathematics at Temple University, which was about machine learning, Eric Albers and I built an MLB Hall of Fame voting predictor using XGBoost. I have since done some updated work on this project (as can be seen in the linked Github), but the attached paper is from our original submission in 2019.
Veering Triangulations and Small Dilatation Pseudo-Anosovs
Part of my PhD work involves doing computations with hyperbolic fibered 3-manifolds to uncover information about small dilatations pseudo-Anosov monodromies thereof. The Github repo linked here displays some of the code. As more about this project gets written down carefully, I will update this page with a write-up explaining the code in more detail.