Queen Mary University of London
Available for placement
I hold a BSc and MSc in Artificial Intelligence at the University of Groningen (NL). During my undergraduate studies, I became captivated by reinforcement learning (RL) agents that learned superhuman gaming capabilities. This sparked my interest in pursuing this direction of AI research. For my thesis, I then successfully created an RL agent with human-level capabilities for the game Lines of Action. During my Master's, my interest shifted toward model-based RL algorithms. For my Master's thesis, I developed multi-task and transfer learning techniques for modern model-based algorithms, resulting in a publication in Transactions on Machine Learning Research. I am now pursuing a Ph.D. at Queen Mary University of London under the guidance of Dr. Paulo Rauber, and am also a Machine Learning Engineer at xDNA.
A description of Remo's research:
Scalable and Efficient Bayesian Algorithms for Reinforcement Learning
My current research focuses on developing RL algorithms that are both scalable and sample-efficient. In particular, the algorithms are based on principled model-based Bayesian algorithms, and I prioritize preserving their core principles in the scalable versions. This is exemplified in my first paper published at International Conference on Machine Learning (ICML), where I successfully scaled the Posterior Sampling for RL algorithm while closely following its original formulation. This resulted in Posterior Sampling for Deep Reinforcement Learning, an algorithm competitive with other state-of-the-art algorithms in Atari games, labeled as a milestone in model-based RL research by one of the conference reviewers.