Service providing system and method based on deep learning

Abstract

The invention belongs to the field of edge data information calculation, and discloses a service providing system and method based on deep learning, so as to protect privacy of training data and improve edge end service quality. In a cloud server side a compression deep learning model of privacy is generated by combining a differential privacy mechanism.A full-connection deep learning model with privacy protection is generated in privacy dense training to protect privacy of training data. In privacy compression training, a full-connection deep learning model with privacy protection is cut. A compression deep learning model with privacy protection is generated, and privacy of training data is protected. The size of the privacy compression model can be reduced to 1 / 9 of the size of the original model, so that the privacy compression model is very suitable for being embedded into an edge server, and edge service quality is improved by accessing the edge server by adjacent mobile equipment.

Description

technical field [0001] The invention belongs to the field of edge data information computing, and in particular relates to a deep learning-based service providing system and method. Background technique [0002] The rapid development of deep learning technology based on neural network has made artificial intelligence (AI) widely used. Deep learning models are widely used in various service delivery systems, including image recognition, speech recognition, and language translation. [0003] The Internet of Things (IoT) was first proposed in 1999 in supply chain management. In recent years, IoT has made significant contributions to healthcare, the environment, and transportation. IoT devices often generate huge amounts of complex data, especially multimedia data. Current neural networks, especially those with multiple hidden layers, are good at processing and analyzing these huge and complex data. Therefore, in IoT applications, deep learning can be used to handle many com...

AI Technical Summary

Problems solved by technology

Compared with traditional privacy protection technology, differential privacy protection can solve two shortcomings of traditional privacy protection technology: on the one hand, it provides an effective and strictly defined mathematical proof that can quantify the level of privacy protection; On the one hand, even if the attacker obtains information other than the target individual object, that is, the maximum background auxiliary information, the attacker cannot deduce the information of the target object

[0011] (1) Existing technologies In the field of artificial intelligence applications of image recognition, speech recognition, and language translation, most IoT applications based on deep learning rely on a centralized cloud architecture, that is, all computing tasks are placed on cloud servers, and deep learning models are deployed. In the cloud server, the limited network bandwidth of the centralized cloud architecture cannot effectively process and analyze these IoT data, especially online learning applications have extremely high real-time requirements, and the centralized cloud server cannot meet the real-time requirements

Because a large number of edge devices generate a large amount of complex data, existing technologies based on centralized cloud architecture cannot effectively analyze and process data, thereby reducing service quality

[0012] Moreover, the existing deep learning model cannot solve the problem of privacy leakage of training data.

[0013] (2) The IoT applications based on deep learning in the existing edge environment lack the consideration of training data privacy leakage, and most of them directly deploy the deep learning model to the server, but the attacker can extract the sensitive information of the training data through the deep learning model, so Will lead to training data privacy leakage

[0014] The difficulty of solving the above technical problems: Due to the limited power, processor speed, data storage capacity and communication resources of edge servers, it is difficult to deploy large-scale deep learning models to edge servers, so it is necessary to compress large-scale deep learning models into small Scale compression of deep learning models, while deploying compressed deep learning models to edge servers will cause training data leakage, so it is necessary to combine differential privacy mechanisms to protect the privacy of training data

However, the higher the privacy level of differential privacy protection training data, the more noise needs to be added in the deep learning process, which will reduce the availability of compressed models, that is, the privacy of training data and the availability of compressed models are contradictory, and the higher the level of privacy protection of training data , the lower the availability of compressed models, so how to balance the privacy of training data and the availability of compressed models is a technical difficulty

Images