Reinforced learning method and device

Abstract

The invention relates to artificial intelligence, provides a reinforced learning method and device, and can improve the training efficiency of reinforcement learning. The method comprises the steps: acquiring a structure chart, wherein the structure chart comprises structure information of an environment or an intelligent agent acquired through learning; inputting a current state and a structure diagram of the environment to a strategy function of the intelligent agent, the strategy function being used for generating an action in response to the current state and the structure diagram, and thestrategy function of the intelligent agent being a diagram neural network; outputting an action to the environment by using the intelligent agent; acquiring a next state and reward data in response to the action from the environment by using the intelligent agent; and carrying out reinforced learning training on the intelligent agent according to the reward data.

Description

technical field [0001] This application relates to the field of artificial intelligence, in particular to a method and device for reinforcement learning. Background technique [0002] Artificial intelligence (AI) is a new technical science that studies theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. Machine learning is at the heart of artificial intelligence. Methods of machine learning include reinforcement learning. [0003] Reinforcement learning is that the agent (agent) learns in a "trial and error" way, and the reward (reward) obtained through the interaction between the action (action) and the environment guides the behavior, and the goal is to make the agent obtain the maximum reward. A policy function is a rule of action used by an agent in reinforcement learning. The policy function is usually a neural network. The strategy function of the agent usually adopts a deep neural network, but the dee...

AI Technical Summary

Problems solved by technology

The strategy function of the agent usually uses a deep neural network, but the deep neural network often has the problem of low learning efficiency

In the case of a large number of training neural network parameters, given limited data or number of training rounds, the expected return of the policy function will be low, so the training efficiency of reinforcement learning is also low

Images