Privacy-Preserving Classification of Personal Data with Fully Homomorphic Encryption: An Application to High-Quality Ionospheric Data Prediction
| Autor: | Maohua Sun, Zheng Li |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Variables
060102 archaeology Computer science media_common.quotation_subject Process (computing) Homomorphic encryption Plaintext 06 humanities and the arts 02 engineering and technology Client-side computer.software_genre Test set Ciphertext 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing 0601 history and archaeology Data mining Protocol (object-oriented programming) computer media_common |
| Zdroj: | Machine Learning for Cyber Security ISBN: 9783030622220 ML4CS (1) |
| DOI: | 10.1007/978-3-030-62223-7_38 |
| Popis: | In recent years, the problem of data leakage is really common, so designing a safe and efficient privacy-preserving machine learning protocol has become an urgent demand to many researchers. The fully homomorphic encryption algorithm has attracted the interest of many researchers. This algorithm can directly operate on the ciphertext without decryption, and the decrypting result is the same as that obtained by performing the same processing on the plaintext. At present, many scholars have designed privacy-preserving machine learning protocols, and have achieved good results. However, as one of the most frequent used algorithms, there are relatively few studies concentrating on the privacy-preserving protocols for logistic regression and fewer focusing on the prediction process. Therefore, we propose a privacy-preserving protocol to solve the data leakage problem during the logistic regression process. Based on the semi-honest assumption and BGV fully homomorphic encryption algorithm, our protocol is mainly focused on the prediction process of logistic regression. In this protocol, there are two parties (client and server), the client holds the test set data and label data, the server holds the model coefficients obtained from the training process. The two parties execute the protocol to finish the logistic regression prediction process in ciphertext. Finally, the client side obtains the prediction category of the dependent variable and calculates the accuracy of the prediction result. During this process, the server will not learn anything about the test set, nor will the client know any information about the model used by the server. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|