apjansen / day4_mx Goto Github PK

I'm Aron Jansen, a physicist now working as a Research Software Engineer at the Netherlands eScience Center.

My research as a physicist was on black holes, with a focus on numerics. Perhaps my biggest contribution was the creation of a Mathematica package (code, paper) to compute oscillations of black holes, and its subsequent use to further our understanding of the (im)possibility of escape from the inside of a black hole (paper).

I now apply machine learning to academic research in various domains:

• NNPDF: Learning what protons are made of
• DEEPDIP: Learned closure models for hydrodynamic simulations with turbulence
• UNSAT3D: Image segmentation of X-ray scans of plant roots in soil

To learn about reinforcement learning I created an AI that can play the ancient board game The Royal Game of Ur. You can learn all about the game and the reinforcement learning techniques used, and more importantly try to beat it, here.

To learn about the concept of equivariance in deep learning, I created a library for group convolutions in Keras. It again has a lot of explanations as well, unfortunately work got in the way of polishing and developing it further.