Model training method based on reinforcement learning and related device

Abstract

The invention discloses a model training method based on reinforcement learning and a related device, which can be applied to a game strategy simulation process. The method comprises the steps of obtaining a preset reinforcement learning model and multiple target reinforcement learning models; inputting the target sample into a preset reinforcement learning model, and performing iterative computation in a reinforcement learning environment to obtain a sample set; extracting an empirical sample from the sample set so as to establish a regularized Anderson target function in combination with thetarget reinforcement learning model; and further determining a loss function by obtaining an Anderson coefficient vector so as to train the preset reinforcement learning model. Samples in the training process are recycled, so that the data utilization rate is improved, the interaction frequency of the intelligent agent and the environment is reduced, and the training efficiency of the reinforcement learning model is improved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a reinforcement learning-based model training method and related devices. Background technique [0002] Reinforcement learning is a mathematical framework for autonomous learning of policies through experience. In recent years, model-free deep reinforcement learning algorithms have been widely used in various challenging fields, such as Atari series of single-player games and multiplayer online battle arena (MOBA) games such as StarCraft. [0003] Generally, for the training process of the reinforcement learning model, in order to obtain a better strategy through training, the agent needs to continuously interact with the environment. [0004] However, for scenes with high-dimensional state spaces such as Go and video games, the number of times the agent needs to interact with the environment is too high, which affects the efficiency of reinforcement learning model t...

AI Technical Summary

Problems solved by technology

[0004] However, for scenes with high-dimensional state spaces such as Go and video games, the number of times the agent needs to interact with the environment is too high, which affects the efficiency of reinforcement learning model training

Images