- Keywords: Model-based Reinforcement Learning, Thompson sampling, Exploration
- Abstract: Posterior Sampling for Reinforcement Learning (PSRL) is a well-known algorithm that augments model-based reinforcement learning (MBRL) algorithms with Thompson sampling. PSRL maintains posterior distributions of the environment transition dynamics and the reward function to procure posterior samples that are used to generate data for training the controller. Maintaining posterior distributions over all possible transition and reward functions for tasks with high dimensional state and action spaces is intractable. Recent works show that dropout used in conjunction with neural networks induce variational distributions that can approximate these posteriors. In this paper, we propose Event-based Variational Distributions for Exploration (EVaDE), variational distributions that are useful for MBRL, especially when the underlying domain is object-based. We leverage the general domain knowledge of object-based domains to design three types of event-based convolutional layers to direct exploration, namely the noisy event interaction layer, the noisy event weighting layer and the noisy event translation layer respectively. These layers rely on Gaussian dropouts and are inserted in between the layers of the deep neural network model to help facilitate variational Thompson sampling. We empirically show the effectiveness of EVaDE equipped Simulated Policy Learning (SimPLe) on a randomly selected suite of Atari games, where the number of agent environment interactions is limited to 100K.
- One-sentence Summary: This paper proposes variational distribution designs to approximate Thompson sampling for model based reinforcement learning in object based domains.
- Supplementary Material: zip