Evaluation of reinforcement learning techniques

摘要

Reinforcement learning is became one of the most important approaches to machine intelligence. Now RL is widely use by different research field as intelligent control, robotics and neuroscience. It provides us possible solution within unknown environment, but at the same time we have to take care of its decision because RL can independently learn without prior knowledge or training and it take decision by learning experience through trial-and-error interaction with its environment. In recent time many research works was done for RL and researchers has also proposed various algorithm and model such as SARSA [2], TDN [3] which tries to solve sequential decision making problems of continuous state and action space.In this paper we proposed Q-learning algorithm and evaluation of RL techniques (Reinforcement learning architecture, algorithms for making training matrix in the form of state-action pair Q-table) containing learner (decision making agent) that takes actions in an environment and receive reward for (or penalty) its actions in trying to solves a problems. Learning agent, the fundamental element of reinforcement learning, there is a decision maker that receive and select an action for the system.In reinforcement learning technique especially in Query base self learning the learner (Agent) required a lot of training input of execution cycle. In order to assess and comparison of QA and TDN based reinforcement learning, we found that QA is better in the context of discount rate, learning time, memory usage.