Sobhan Mohammadpour

I’m a master’s student at Université de Montréal (MILA and CIRRELT), advised by Emma Frejinger and Pierre-Luc Bacon. I work on end-to-end or decision aware learning. I try to find the real problem we should be optimizing not the auxiliary loss that is convenient. Concretely I’ve worked on learning the features we need to do inverse reinforcement learning. At the moment I’m trying to define a gradient for the shortest path, and I’m working on finding a better smooth Bellman operator for reinforcement learning.