Deep Learning Training Management Platform Based on Distributed Technologies in Resource-Constrained Scenarios

摘要

Deep learning has attracted a lot of research attention in the past few years for its efficiency and accuracy. However, there exist two problems of their study. Firstly, the computing power of one single machine is limited and not suitable for handling with training deep learning models with massive cells. Secondly, it costs much to train models on different deep learning frameworks. Motivated by these problems, this paper proposed a deep learning training management platform based on distributed technologies, which integrates different kinds of deep learning frameworks through virtualization technologies and coordinates machines through distributed technologies. Specially, specific algorithms are proposed to solve the multi-task scheduling problem, the computing resources allocation problem and the fault tolerance problem in resource limited scenarios. It turns out that the platform can be widely used in small and medium-sized research teams.