Théo graduated from the University of Bordeaux, France, in 2017 with a BS in physics and chemistry and a minor in green and bio-sourced chemistry. He completed his last year at Northern Arizona University in Flagstaff, AZ, where he worked on simulations of magnetic interactions between self-propelled nanoparticles. He is now completing a master’s program in physical chemistry and chemical physics in Bordeaux, where he’s specializing in theoretical and numerical methods for material science. Théo likes computational chemistry so much, he conducted an independent study at the ISM (Institute for Molecular Science) where he coded a reaction dynamics simulation tool and then built a small computation cluster with Raspberry Pis, which he’s trying to upgrade as a hobby.
In his spare time, Théo enjoys hiking in the woods, coding stuff, cooking french cuisine and reading science-fiction and philosophy books. He has strong interests in history, psychology, political science and neurology. As a physics enthusiast, he was elected social outreach officer for the Society of Physics Students club at NAU, where he organized fundraisers, physics demonstrations to elementary school kids and talked about chemistry. He used to be captain of his dramatic improvisation team in high school and college and still enjoys practicing theater.
Artificial Life and Cellular Automata: Langton’s Ant »
Project: Pixel Segmentation for Feature Extraction from Satellite Pictures with Neural Networks
Goal of the project:
I wanted to extract basic topographic features like roads and highways from high-resolution satellite pictures using a neural network.
Summary of work:
I first tested the freshly implemented GeoImage function to obtain the satellite pictures, then automated the generation of matching sets of binary masks for the learning process. The next step was the implementation of the experimental neural network eNet in Mathematica, followed by an intensive training. Multiple problems emerged during the setup of this workflow, which we will discuss later.
Results and future work:
I have in mind a few future augmentations/ameliorations of the model. A possible application would be the automatic update of the GPS databases.