Big data technologies, such as machine learning, have increased data utility exponentially. At the same time, the cloud has made the deployment of these technologies more accessible. However, computations of unencrypted sensitive data in a cloud environment may expose threats and Cybersecurity attacks. We consider a class of innovative cryptographic techniques called Privacy-Preserving Technologies (PPTs) to address this problem. That might help increase utility by taking more significant advantage of the cloud and machine learning technologies while preserving privacy.The first section provides a brief introduction to the so-called Homomorphic Encryption "HE" by giving an overview of the most promising schemes and then giving the current state of the art of HE tools such as Libraries and Compilers. This section aims to help non-cryptographer developers propose HE solutions by explaining what makes developing HE applications challenging.Then, we address the Privacy-Preserving in Machine Learning (PPML), an approach that allows to train and deploy ML models without exposing their private data. After exploring state-of-the-art for the most used ML models in PPML, we will overview applications of Homomorphic Encryption in Machine Learning.
Machine learning as a service" (MLaaS) in the cloud accelerates the adoption of machine learning techniques. Nevertheless, the externalization of data on the cloud raises a serious vulnerability issue because it requires disclosing private data to the cloud provider. This paper deals with this problem and brings a solution for the K-nearest neighbors (k-NN) algorithm with a homomorphic encryption scheme (called TFHE) by operating on end-to-end encrypted data while preserving privacy. The proposed solution addresses all stages of k-NN algorithm with fully encrypted data, including the majority vote for the class-label assignment. Unlike existing techniques, our solution does not require intermediate interactions between the server and the client when executing the classification task. Our algorithm has been assessed with quantitative variables and has demonstrated its efficiency on large and relevant real-world data sets while scaling well across different parameters on simulated data.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.