Artificial intelligence reinforcement learning service platform

Abstract

The invention relates to an artificial intelligence reinforcement learning service platform. The system is structurally divided into an infrastructure layer, an application service layer and an interface access layer, the infrastructure layer provides network resources, computing resources, storage resources and virtualization service resources required by the reinforcement learning service platform, and provides IT infrastructure services related to cloud storage and cloud processing through virtualization, load balancing, disaster recovery backup and elastic computing technologies; the application service layer comprises a Project-based packaging and management module, a cloud development and debugging environment module and a virtual development environment interface module; the three systems provide various universal or self-defined cloud research environments for field researchers from top to bottom; the platform encapsulates data, algorithms and research environments involved inreinforcement learning research in a virtualization container in a Project form, and opens up an independent test environment for each user using the platform; and the interface access layer can enable the reinforcement learning researcher to manage the cloud computing environment in a self-service manner.

Description

technical field [0001] The invention relates to a pre-reinforcement learning development platform technology, in particular to an artificial intelligence reinforcement learning service platform. Background technique [0002] Machine learning is one of the core issues of artificial intelligence, which is to study and simulate human learning behavior, and to generate new knowledge through learning after acquiring knowledge. Data-based machine learning is one of the important methods in modern intelligent technology. Research starts from observed data (samples) to find laws and acquire knowledge, and use these laws and knowledge to predict future data or unobservable data through a certain learning mode. . Machine learning can be classified into supervised learning, unsupervised learning and reinforcement learning according to the learning mode. The goal of reinforcement learning is to learn the mapping from environmental state to behavior, so that the behavior selected by th...

AI Technical Summary

Problems solved by technology

Although these environments provide a standardized research environment in specific fields to a certain extent, the development and research based on these reinforcement learning environments still face the following problems: training reinforcement learning algorithms requires a large One-stop scientific research environment; it takes a lot of time to deploy the corresponding reinforcement learning development environment, and it is difficult to reproduce the algorithm due to different software versions and hyperparameters; the server side lacks visual development tools, and cannot observe the reinforcement learning agent environment in real time Simulation, difficult to quickly develop and verify code

Images