Abstract:As one of the most popular queries in big data era, the $k$ nearest neighbors ($k$NN) query plays a significant role in various applications, such as medical diagnosis, signal processing, and recommendation systems. Meanwhile, driven by the advancement of the cloud service, an emerging trend among applications is to outsource the dataset and the corresponding $k$NN query services to the cloud. However, as the cloud is not fully trusted, those applications will face vital privacy concerns, and thus they usually encrypt data before outsourcing them to the cloud. Because encrypted data are outsourced to cloud, the $k$NN query over encrypted data has become increasingly attractive, and many solutions have been put forth in recent years. However, existing solutions cannot fully satisfy the objects of returning exact query results, protecting database privacy and query privacy, achieving high query efficiency, and imposing low computational costs at the user side. To address these issues, in this paper, we propose a new practical and privacy-preserving $k$NN query scheme. Specifically, we first refine the general security requirements for the matrix encryption by systematically analyzing existing algorithms. Then, we design a novel asymmetric matrix encryption (AME) to securely achieve Euclidean distance computation and two distances comparison in a single-party and non-interactive way. Then, based on the AME scheme, we propose a privacy-preserving $k$NN query scheme, in which a max-heap of size $k$ is used to accelerate query efficiency. Detailed security analysis shows that our proposed scheme is really privacy-preserving. In addition, extensive performance evaluations are conducted, and the results demonstrate that our proposed scheme is also highly efficient.

Insecurity and Hardness of Nearest Neighbor Queries Over Encrypted Data

Adaptive Secure Nearest Neighbor Query Processing Over Encrypted Data

Approximate Message Passing with Nearest Neighbor Sparsity Pattern Learning

Industrial Scale Privacy Preserving Deep Neural Network

Achieving Practical and Privacy-Preserving kNN Query over Encrypted Data

Ciphertext-Only Attack on a Secure $k$-NN Computation on Cloud

Secure K-Nn Query on Encrypted Cloud Data with Multiple Keys

Secure Computation over Encrypted Databases

Secure State Estimation for Artificial Neural Networks With Unknown-But-Bounded Noises: A Homomorphic Encryption Scheme

Towards Scalable and Privacy-Preserving Deep Neural Network via Algorithmic-Cryptographic Co-design

Secure k-Nearest Neighbor Query over Encrypted Data in Outsourced Environments

Toward Scalable and Privacy-preserving Deep Neural Network Via Algorithmic-Cryptographic Co-design

Time-Restricted, Verifiable, and Efficient Query Processing over Encrypted Data on Cloud

Privacy-Preserving Queries On Encrypted Data

Enabling Secure In-Memory Neural Network Computing by Sparse Fast Gradient Encryption

Interpreting and Mitigating Leakage-Abuse Attacks in Searchable Symmetric Encryption

Performance Enhanced Secure Spatial Keyword Similarity Query With Arbitrary Spatial Ranges

A Homomorphic Encryption Framework for Privacy-Preserving Spiking Neural Networks

A secure kNN query processing algorithm using homomorphic encryption on outsourced database

Interpretable Encrypted Searchable Neural Networks

SparseLock: Securing Neural Network Models in Deep Learning Accelerators