Distributed deep learning classification method based on alternating direction multiplier method ADMM

Abstract

The invention provides a distributed deep learning classification method based on an alternating direction multiplier method (ADMM). The distributed deep learning classification method comprises the following steps of 1, performing classification marking on images, videos and files in a database of each node; 2, initializing Alexnet network layer parameters of each node and a Lagrange multiplier matrix; 3, extracting features of data of each node through one-time forward propagation of an Alexnet network; 4, introducing a minimized classification error to obtain a globally optimal classifier;5, assigning a global classifier parameter to the last layer, namely a full connection layer, of each node Alexnet network, and performing forward propagation on the data of each node again through the Alexnet network; 6, fixing classification layer parameters, and updating feature layer parameters; 7, judging whether the training precision is equal to 1 or not, if yes, completing training, and otherwise, repeating the steps 3-7. According to the method, the problems that the number of samples such as images, videos and texts is large, and if the samples are concentrated together for training,the transmission amount and the calculation amount are too large are solved.

Description

technical field [0001] The invention relates to a deep learning method, specifically a distributed deep learning classification method, which belongs to the technical field of machine learning. Background technique [0002] With the continuous development of social networks, e-commerce, mobile Internet, etc., the scale of data storage and processing is getting larger and larger, and stand-alone systems can no longer meet the growing needs. Internet companies such as Google and Alibaba have successfully spawned the two hot fields of cloud computing and big data. Both cloud computing and big data are applications built on distributed storage. The core of cloud storage is the back-end large-scale distributed storage system. Big data not only needs to store massive amounts of data, but also needs to analyze these data through appropriate frameworks and tools to obtain useful parts. If there is no distributed storage, it will Not to mention the analysis of big data. Although th...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to provide a distributed deep learning classification method based on the Alternating Direction Multiplier Method ADMM, to solve the problem that the number of samples such as images, videos, and texts is large, and if they are trained together, the transmission volume and calculation volume are too large. Through distributed training, the ADMM method is used to obtain a globally optimized classifier, and on the basis of the global classifier, the BP algorithm is used to obtain a feature layer suitable for the global classifier, while protecting the data independence of each node in the distributed training, and Classify the samples

Images