Abstract

     In recent years, deep learning (DL) has been ubiquitous in several areas, such as text recognition and data analysis, limited by this and increasingly used in security and data protection applications. Thus, the DL method has achieved remarkable big data analysis growth to avoid different attacks. This paper presents different methods for protecting privacy for DL in big data analysis. First, some possible attacks are explained, and then some basic approaches to protecting privacy in big data platforms are explained. In each section, drawbacks of the corresponding attacks are elaborated, and DL-based methods’ effectiveness in privacy preservation has been discussed. Finally, an effective solution for enhancing privacy preservation in DL models is given. The several DL-based privacy preservation methods for big data analysis and their advantages and disadvantages are elaborated. At last, drawbacks of DL based methods are highlighted, and future scope is given to address these issues.

Introduction

     In the era of digital technology, incredible amounts of data are produced by multiple organizations such as social media, banks, end sources and hospitals, etc. Social media generates tons of data every day, leading to huge data.1 Big data refer to the data collected from various sources such as machines, people, and things. The human-derived data can be individuals, text generation, and videos uploaded to the Internet. Machines generate various files, multimedia, and audits, while the data collected by multiple digital sensor devices are called data of things.2,3 Characteristics of big data are expressed by 5 V; large volume, higher velocity, greater variety, high value, and lower veracity. Volume denotes the speed of data collected, while the type of collected data is called diversity.4 Technological advances in healthcare make it easier to collect patient data electronically, which is presented as big data.5 In the modern medical system, patients are treated with multiple medical records. Thus, it is necessary to confirm secure data exchange to facilitate patient treatment in multiple hospitals.6

Conclusion

     The technological advancements used the digital platform for data communication simultaneously, and data privacy is a prime concern. The data associated with several fields must be processed securely with information leakage. For analyzing a large amount of data, the DL methods are adopted. At the same time, several attacks related to these methods damage the data communication, especially in the case of the big data platform. Thus, DL privacy preservation has become an important research area due to the privacy concern of a large amount of private data. If an attacker accesses personal information, it will cause data loss to users. Moreover, the information leakage in DL is happened due to internal and external factors. Thus, an effective approach toward privacy protection schemes greatly influences the enhancement of privacy preservation in DL. Several privacy preservation methods are reviewed in this work and their pros and cons. The methods reviewed in this paper are classified based on anonymization, optimizationbased approaches, and cryptographic methods. Moreover, several possible attacks related to DL were recalled. We can conclude from these papers that DL-based methods are more effective than other classic methods.