1.6 Summary
Reinforcement learning is a computational method for understanding and automating goal-oriented learning and decision-making. The difference between it and other computing methods is that it emphasizes that an agent learns from direct interaction with its environment, without the need for exemplary supervision or a complete model of the environment. In our opinion, reinforcement learning is the first field to seriously solve the computational problems arising from interactive learning with the environment in order to achieve long-term goals.
Reinforcement learning uses the formal framework of the Markov decision process to define the interaction between the learning agent and the environment, including state, action and reward. This framework aims to be a simple way to express the essential characteristics of artificial intelligence problems. These characteristics include a sense of cause and effect, uncertainty and uncertainty, and the existence of clear goals.
The concept of value and value function is the key to most of the reinforcement learning methods we consider in this book. We believe that in the policy space, the value function is important for effective search. The use of value function distinguishes reinforcement learning methods from evolutionary methods, which directly search the strategy space under the guidance of the entire strategy evaluation.