Advances in Value-based, Policy-based, and Deep Learning-based Reinforcement Learning

Abstract

Machine learning is a branch of artificial intelligence in which computers use data to teach themselves and improve their problem-solving abilities. In this case, learning is the process by which computers use data and algorithms to build models that improve performance, and it can be divided into supervised learning, unsupervised learning, and reinforcement learning. Among them, reinforcement learning is a learning method in which AI interacts with the environment and finds the optimal strategy through actions, and it means that AI takes certain actions and learns based on the feedback it receives from the environment. In other words, reinforcement learning is a learning algorithm that allows AI to learn by itself and determine the optimal action for the situation by learning to find patterns hidden in a large amount of data collected through trial and error. In this study, we introduce the main reinforcement learning algorithms: value-based algorithms, policy gradient-based reinforcement learning, reinforcement learning with intrinsic rewards, and deep learning-based reinforcement learning. Reinforcement learning is a technology that enables AI to develop its own problem-solving capabilities, and it has recently gained attention among AI learning methods as the usefulness of the algorithms in various industries has become more widely known. In recent years, reinforcement learning has made rapid progress and achieved remarkable results in a variety of fields. Based on these achievements, reinforcement learning has the potential to positively transform human lives. In the future, more advanced forms of reinforcement learning with enhanced interaction with the environment need to be developed.