Most Recent Project
Terrain Diffusion
A diffusion model for generating realistic terrain using a U-Net architecture. Trained on heightmaps of real locations, it allows to directly generate a surface with river valleys and other erosion features without the need to run an expensive erosion algorithm. Full article describing the model and how it works is available on LinkedIn.
About Me
I started programming in middle school with C as my first language. I leveraged it to win a number of programming competitions in highschool and used it for bioinformatics and biochemistry research throughout undergrad at Dalhousie University .My post-graduate studies focused on molecular dynamics (MD): developing methods and software to efficiently model polarization in aqueous and biological media. I transitioned to primarily using Python as my language of choice and used it to develop atomic resolution water models valid across a wide range of temperatures. This work culminated in a PhD from TU/E in 2019.
During my PhD I taught myself ML techniques both as an aid in optimizing MD models and for personal projects. This started as applications of clustering and dimensionality reduction to simulation trajectories and ended up as a deep fascination with neural networks. Expertise I gained this way allowed me to later create a completely in-silico approach for ligand optimization driven by regression models. It allows one to optimize drug candidates without needing to synthesize and test the intermediate guesses in a lab, relying instead on free energy calculations as a highly accurate oracle and active learning for limiting the number of necessary ligand evaluations.
In my free time I build ML models of commodity prices, develop games and procedural asset generators, and run two Dungeons and Dragons campaigns.
If you would like to learn more about me and what I do, take a look at some of my projects.