I work in reinforcement learning, which can be loosely described as reward-driven adaptive decision-making. Not only is this a rich and intellectually stimulating area, but it is also on the verge of impacting a wide array of industrial and consumer technologies. My general goal is to be on top of the theory and algorithms, while doing my best to guide the application towards directions that are, as far as I am able to tell, beneficial to society.
My PhD work focused on investigating the properties of value function estimators in batch reinforcement learning. This involves using mathematics and simulations in order to analyze methods for predicting the total reward obtained under some intervention policy. The math is mostly statistics (looking at estimator biases, variances, and consistency), and the simulations have been based on natural resource management models (one of the Alaskan halibut, and another of the Canadian mallard).
Currently, I am involved in the management of the natural resources inside the Earth's crust - I recently started a postdoc in mining engineering as part of McGill's COSMO Lab. This is proving to be a good opportunity to learn more about modelling systems where humans and nature interact, and to contribute to better mining practices in the process.
I started my PhD at McGill's Reasoning and Learning Lab in 2007, under the supervision of Doina Precup. A few years after, Joelle Pineau has become my co-supervisor. They are a great team to work with. Previously, I completed an MSc in computer science at the University of Alberta under the supervision of Rich Sutton and Vadim Bulitko, and a BSc (also in computer science, but with a solid dose of mathematics) from the University of Bucharest.