UMass Machine Learning and Friends Lunch | Main / An Object-oriented Representation For Efficient Reinforcement Learning

Rich representations in reinforcement learning have been studied for the purpose of enabling generalization and making learning feasible in large state spaces. I will introduce Object-Oriented MDPs (OO-MDPs), a representation based on objects and their interactions, which is a natural way of modeling environments and offers important generalization opportunities. I will discuss a few algorithms for learning OO-MDPs, their advantages and shortcomings. I illustrate the performance gains of our representation and algorithm in the well-known Taxi domain, plus a real-life Atari videogame (Pitfall!).