Research On Distributed Deep Learning Task Assignment Algorithm Based On Blockchain

As the hardware performance of mobile devices improves,it has become an inevitable trend to improve user experience by deep learning and neural network.Considering the mobility and portability of mobile devices,training a deep learning model should reduce the delay of tasks and the power consumption of devices.In order to speed up the training while ensuring privacy and security,there are already some projects for distributed training of mobile devices on the blockchain.Among them,the parallel distributed training of mobile devices greatly shortens the delay of training.The blockchain network records the process of distributed training and ensures the security of the system through the tamper-resistant manner.In order to reduce energy consumption and the response time of the system while improving the throughput,an innovative distributed training task allocation framework for mobile devices is proposed.The framework which optimizes the energy efficiency of the system through user selection and resource allocation includes a task evaluation standard that comprehensively considers the difficulty of the task and user attributes.Besides,task allocation framework in which blockchain is used to motivate users to provide interactive data to increase the accuracy is proposed.The framework classifies tasks to improve the utilization rate of various mobile devices with the objective of reducing the energy consumption of executing tasks and improving the overall revenue under the QoS constraint.Then the high-dimensional state space control problem in dynamic environment is solved by deep reinforcement learning.Finally,it shows the simulation results under different parameters which proves the feasibility of the proposed framework.