Portfolio

Improve the reliability of tractography methods by reinforcement learning by using an oracle that quantifies the anatomical validity of a white matter (WM) fiber. We generate randomly new trajectories of a possible fiber from a failling fiber and select the best trajectory based on the oracle’s validity score.

An attempt to test state-of-the-art reinforcement learning algorithms in a multi-agent cooperative and competitive setting using ML Agents SoccerTwos.

An Open Source framework to train reinforcement learning agents on the Tetris NES environment 100% written in Julia.