BEERL: Both Ends Explanations for Reinforcement Learning

Deep Reinforcement Learning (RL) is a black-box method and hard to understand because the agent employs neural network (NN). To explain behavior decisions made by agent, different eXplainable RL (XRL) methods are developed; for example, feature importance applied analyze contribution of input side model, reward decomposition components output end model. In this study, we present novel connect explanations from both ends which results in fine-grained explanations. Our exposes prioritization user, turn generates two levels explanation allows reconfigurations when unwanted behaviors observed. The further summarizes detailed into focus value that takes account all quantifies fulfillment desired properties. We evaluated our applying it remote electrical telecom-antenna-tilt use case openAI gym environments: lunar lander cartpole. demonstrated detailing features’ contributions certain rewards revealed biases components, then addressed adjusting reward’s weights.