Deep reinforcement learning method based on multiple threads

• Main Authors: FAN YINGBAO, WANG XIAO, RUAN AIWU
• Format: Patent
• Language: Chinese; English
• Published: 18.07.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; PHYSICS

The invention provides a deep reinforcement learning method based on multiple threads, and belongs to the technical field of machine learning. In a deep reinforcement learning algorithm, training data is obtained through continuous interaction between an intelligent agent and an environment, and the process needs to consume a large amount of time to obtain enough data. According to the method, the data sample collection speed is increased in a multi-thread synchronous sampling mode, specifically, the overall algorithm is divided into a sample collection part and a network training part, and in the sample collection part, intelligent agent synchronization in multiple sub-threads interacts with the environment to generate data; and the parameter training part is used for training and updating network parameters in the main thread by utilizing the data acquired in the sub-thread. The sub-thread is only responsible for sample collection, and the main thread is only responsible for network training. In this way, t