Security Research Based On Three Typical Machine Learning Algorithms

The combination of big data processing and cloud is inevitable.Big data processing requires powerful computing power and storage space,and the dynamic allocation of cloud resources just meets this demand.The resource advantage of cloud can provide a suitable platform for big data processing.However,the application scenarios of the combination of big data and cloud need to solve the problem of data privacy preserving.An open cloud environment not only needs to face security threats and challenges from inside and outside,but also semi-trusted and honest-but-curious cloud services make the problem even more difficult.How to find a processing method,which can not only effectively ensure data privacy and security during big data processing,but also make full use of the incompletely trusted cloud environment to solve the problem has become the key to the combination of big data processing and cloud.In this thesis,from the perspective of data privacy preserving,combined with the machine learning algorithms running in the cloud required for data processing,a privacy preserving model is proposed,which expands the application scenario of combining big data and cloud,and provides beneficial ideas and solutions for data privacy preserving.The main contents of this thesis are as follows:In order to apply the classical supervised learning algorithm logistic regression algorithm to the data processing process,a privacy-preserving logistic regression algorithm is proposed.The algorithm can ensure the privacy of the data not to be disclosed while executing the logistic regression algorithm in the cloud.The entire calculation in the cloud is carried out in an encrypted environment,which can completely ensure the privacy of the data.Improve computing efficiency by taking advantage of the powerful computing power in the cloud.In order to solve the privacy risks of the widely used deep learning algorithms when processing data,a privacy preserving deep learning algorithm is proposed.Like the privacy preserving logistic regression algorithm,the main calculation of the privacy preserving deep learning algorithm is also performed in the cloud,and the encryption and decryption operations are placed on the local side.The computing in the cloud is performed in an encrypted state through homomorphic encryption to ensure data privacy.In order to strengthen the protection of data privacy information and ensure the security in the process of multi-party interaction,based on k-means clustering algorithm,combined with secure multi-party computation,a privacy preserving k-means clustering algorithm related to secure multi-party computation is proposed.Both the privacy preserving logistic regression algorithm and privacy preserving deep learning algorithm only involve the interaction between the local side and the cloud,while the privacy preserving k-means clustering algorithm involves multi-party interaction and needs to ensure the data privacy of each party.This research provides three floor-to-ceiling privacy processing algorithms for data privacy preserving,provides theoretic basis and useful exploration for understanding privacy preserving in the overall process of data processing,can expand the application scenarios of data security processing,and has clear problem orientation and practical value.