State of Art of Decision Making Methods Based on Reinforcement Learning

With the advent and development of the era of big data, machine learning and artificial intelligence play a key role in high-risk fields such as credit, medical care, and law, becoming a powerful decision support tool for improving efficiency and identifying risks. Reinforcement learning mimics the decision-making process of human brain, which is consistent with the empirical thinking of human beings when solving practical problems, so it is applied to the decision-making of simulating human brain thinking. In this paper, reinforcement learning decision making methods are reviewed based on the actual decision-making needs of the industry. This paper introduces the basic theories of reinforcement learning such as Markov decision process, value function and policy optimization. The current mainstream reinforcement learning algorithms can be divided into model-based methods and model-free methods according to whether the agent uses the environment model or not. The latter includes both online and offline learning. In view of the difficulty of establishing accurate models, this paper focuses on model-free methods. In addition, meta-reinforcement learning that can quickly adapt to new tasks, which may be model-free or model-based, is also introduced. Finally, this paper provides a macro perspective for the in-depth study of decision-making methods based on reinforcement learning, and provides strategic suggestions for solving practical problems.