Joint learning method for features and strategies based on state features and subsequent features

Abstract

The invention discloses a joint learning method for features and strategies based on state features and subsequent features. The method comprises: obtaining state features representing an input stateby learning mapping from the input state to the instant award; by learning mapping from state features to a value evaluation function, obtaining subsequent features; the obtained state features and the subsequent features being on different time resolution ratios, after the state features and subsequent features are fused, learning fusion results by using a variety of strategic learning networks.Compared with a traditional Agent network, the method uses sample information more efficiently. Compared with other algorithms, learning speed is obviously accelerated, and a network can converge faster and obtain better learning effect.

Description

technical field [0001] The invention relates to the field of deep reinforcement learning, in particular to a joint learning method of features and strategies based on state features and subsequent features. Background technique [0002] Deep Reinforcement Learning (Deep Reinforcement Learning) is a sequential decision-making learning method based on deep networks. It integrates deep learning and reinforcement learning, realizes end-to-end learning from state to action, and continuously interacts with the environment. Implement policy enhancements in . In high-dimensional complex problems, effective features are automatically extracted from perceptual information based on deep neural networks, and based on this, policy learning is performed and actions are directly output, that is, there is no hard-coded process in policy learning. Deep reinforcement learning can effectively solve the perceptual decision-making problems of agents (Agents) under high-dimensional complex probl...

AI Technical Summary

Problems solved by technology

On the one hand, low-quality samples will affect the training and convergence of the network. On the other hand, the network training process does not have a high utilization rate of samples; these problems greatly increase the training cost of the network.

Images