Patrick Lehman

Background

About Patrick Lehman

UF Computer Engineering student. In my free time I enjoy cooking, reading sci-fi (especially Neal Stephenson) and CrossFit. My research interests center around novel variations of Vision Transformers, Neural ODE's and Heuristic Machine Learning (NN+ MCTS, Environment Modelling, etc). Below you can read a synopsis of my career thus far, in reverse chronological order: As an Algorithm Engineering Intern at Neurava, I automated energy-entropy ratio calculations, improving feature capture by 358%. I utilized advanced techniques like Continuous Transformers, Decision Trees, and Wavelet Transforms for analyzing biological signals and developed an automated pipeline for epilepsy detection algorithms. As a Machine Learning Research Intern at Fire Neural Network, I optimized image segmentation using a Vision Transformer. Leveraging PyTorch Lightning and Weights and Biases, I created an automated pipeline, integrated satellite images with geoJSON data, and achieved a 4.5x increase in model throughput. As a Research Assistant at the University of Florida Interactive Visualization and 3D Graphics Lab, I trained various segmentation models, including Video Vision Transformers. Utilizing Lightning AI and distributed GPUs accelerated the training process. In the University of Florida Cognition and Decision Modeling Lab, I developed models for poorly characterized distributions, conducted analysis on Bayesian Computation and Normalizing Flows, and translated statistical models between MATLAB and Python. As a Software Engineer Intern at IT Assist Team, I contributed to an API for charity functions, integrated third-party services, tested their security, and published over lines of code. Additionally, I've worked on some fun projects: Educatus, a NextJS web app offering AI-based study utilities; medmodeler, a PyTorch Vision Transformer for CAT Scan segmentation; and the website Rollsolid, a data-based utility for Texas Holdem Poker.