Blockchain driven secure and private machine learning algorithms for post-quantum 5G/6G enabled industrial IOT with applications to cyber-security and health

Warning The system is temporarily closed to updates for reporting purpose.

Kjamilji, Artrim (2021) Blockchain driven secure and private machine learning algorithms for post-quantum 5G/6G enabled industrial IOT with applications to cyber-security and health. [Thesis]

[thumbnail of 10387049.pdf] PDF
10387049.pdf

Download (6MB)

Abstract

We provide a general framework for secure and private multi-label multi-output machine learning (ML) algorithms for the semi-honest model in distributed edge IoT (Internet of Things) environments enabled by 5G/6G networks. The proposed framework includes the special cases of binary, multi-class and multi-label ML algorithms. We deal with both horizontally and vertically partitioned datasets. Initially, (i) we propose novel secure feature selection protocols by homomorphically evaluating features’ information gains in distributed environments, we proceed with (ii) novel secure training protocols over the set of selected features, then (iii) we propose novel secure building blocks which are commonly used on ML algorithms (e.g. secure sum, comparison, argmax, top-K, sorting, permutation, etc.), as well as on secure linear algebra (e.g. secure inner product, cascading matrix-vector and matrix-matrix multiplications, matrix transpose, etc.), and finally (iv) on top of proposed secure building blocks we build our novel secure ML classification protocols for various ML classifiers such as Deep Neural Networks (DNN), Support Vector Machines (SVM), Decision Trees (DT) and Random Forests (RF), different flavors of Naïve Bayes (NB), Logistic Regression (LR) and K Nearest Neighbors (KNN). Moreover, our secure classification protocols also deal with malicious users that arbitrarily deviate from the protocol and they show no loss of accuracy due to secure classifications. In the process, our participants interact with each other in order to fulfill strict security. privacy and efficiency requirements. To these ends, we provide confidentiality, integrity and authenticity to each interaction by signing their hashed contents with the corresponding participants’ private key. We assure the consistency among interactions by introducing timestamps and linking them with the hashed content(s) of the preceding interaction(s). This makes our protocols a natural fit for blockchain technology. Moreover, the proposed cryptographic tools are proven to be resistant to quantum computer attacks, making our protocols applicable to the post quantum world. We did our theoretical analysis and extensive experimental evaluations over benchmark datasets related to cyber-security and health. They show that our protocols have an advantage ranging from several times to orders of magnitudes with respect to the state-of-the-art in terms of computation and communication costs. This makes our protocols among the most efficient ones in literature. Also, they are among the best in terms of security and privacy properties and allow high rate of fault tolerance and collusion attacks of dataset owners with respect to the state-of-the-art.
Item Type: Thesis
Uncontrolled Keywords: blockchain. -- multi-label multi-output machine learning algorithms. -- secure IoT. -- privacy preserving. -- feature selection. -- training. -- classification. -- homomorphic encryption. -- collusion attacks .-- distributed environments .-- cyber-security. -- Internet of Medical Things. -- blok zincir.-- çok etiketli çok çıktılı makine öğrenimi algoritmaları. -- güvenli nesnelerin internet. -- mahremiyet koruma. -- öznitelik seçimi. -- eğitim. -- sınıflandırma. -- homomorfik şifreleme. -- gizli anlaşma saldırıları. -- dağıtık ortamlar. -- siber güvenlik. -- Tıbbi Nesnelerin İnterneti.
Subjects: T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7800-8360 Electronics > TK7885-7895 Computer engineering. Computer hardware
Divisions: Faculty of Engineering and Natural Sciences > Academic programs > Computer Science & Eng.
Faculty of Engineering and Natural Sciences
Depositing User: IC-Cataloging
Date Deposited: 19 Oct 2021 10:01
Last Modified: 26 Apr 2022 10:38
URI: https://research.sabanciuniv.edu/id/eprint/42493

Actions (login required)

View Item
View Item