PhD student, Montefiore Institute, Department of Electrical Engineering and Computer Science, ULiège, Belgium
Deep Quality-Value (DQV) Learning
I will argue for the importance of jointly approximating the state-value function alongside the state-action value function in model-free Deep Reinforcement Learning (DRL). I will introduce a novel family of DRL algorithms which follow this specific learning dynamic, and will show how these algorithms learn significantly faster and better than algorithms which only approximate the state-action value function.
Matthia Sabatelli is a PhD student at the Department of Electrical Engineering and Computer Science of the Université de Liège, Belgium under the supervision of Dr. Pierre Geurts. Before joining Belgium he received his Msc. Artificial Intelligence from the University of Groningen. His research interests currently lie at the intersection between Supervised and Deep Reinforcement Learning with a particular focus on Transfer Learning.