Self-adaptive resource management method and device for distributed reinforcement learning training

Abstract

The embodiment of the invention provides a self-adaptive resource management method and device for distributed reinforcement learning training. The self-adaptive resource management method comprises the steps: giving a resource demand initial value to a newly-added training task according to task information when the newly-added training task is submitted; counting current distributed cluster resource supply residues, and judging whether new resources need to be derived or not; if so, determining the number and configuration of newly-added virtual machines, adding the newly-added virtual machines into the distributed cluster, and arranging newly-added training tasks according to a preset task arrangement process; and if not, arranging the newly-added training tasks according to the presettask arrangement process. For the self-adaptive resource management method and device for distributed reinforcement learning training, after the newly added training task is received, resource derivation is carried out according to the remaining condition of the distributed cluster resources, and then task arrangement is carried out or directly carried out, so that automatic deployment of the training task is realized, and the resource use cost and the labor cost of distributed reinforcement learning are remarkably reduced.

Description

technical field [0001] The invention relates to the technical field of cloud computing and distributed reinforcement learning, in particular to an adaptive resource management method and device for distributed reinforcement learning training. Background technique [0002] Reinforcement learning is a general term for a class of machine learning algorithms, and together with supervised learning and unsupervised learning, it constitutes the three major branches of machine learning. The training process of reinforcement learning is a sequential decision-making problem, which studies how the agent acts based on the feedback of the environment to maximize the expected benefits. A multi-agent system consists of a group of autonomous, interactive entities that share the same environment, perceive the environment through perceptrons and take actions through actuators. Training agents in a multi-agent system through reinforcement learning technology can effectively improve their over...

AI Technical Summary

Problems solved by technology

Once the system falls into an under-provisioning state, the stability and performance of the training process cannot be effectively guaranteed

Images