Abstract:Homomorphic encryption (HE) is pivotal for secure computation on encrypted data, crucial in privacy-preserving data analysis. However, efficiently processing high-dimensional data in HE, especially for machine learning and statistical (ML/STAT) algorithms, poses a challenge. In this paper, we present an effective acceleration method using the kernel method for HE schemes, enhancing time performance in ML/STAT algorithms within encrypted domains. This technique, independent of underlying HE mechanisms and complementing existing optimizations, notably reduces costly HE multiplications, offering near constant time complexity relative to data dimension. Aimed at accessibility, this method is tailored for data scientists and developers with limited cryptography background, facilitating advanced data analysis in secure environments.

What problem does this paper attempt to address?

This paper attempts to solve the efficiency problems encountered by homomorphic encryption (HE) when processing high - dimensional data, especially in machine learning and statistics (ML/STAT) algorithms. Specifically, although homomorphic encryption can achieve secure computation on encrypted data, it faces huge computational challenges when dealing with complex non - linear data analysis, especially operations involving internal non - linear functions. Traditional linear models such as linear regression are suitable for HE, but when extended to more complex models (such as logistic regression), they often lead to over - simplified linear or quasi - linear methods, thus limiting the application of HE in complex models. To solve these problems, the author proposes an effective acceleration technique based on the kernel method to optimize the time performance in the HE scheme. This method is independent of the underlying HE mechanism and can work in synergy with existing optimization techniques, significantly reducing expensive HE multiplication operations and providing a time complexity close to constant with respect to the data dimension. In addition, this method is designed to be easy to use, targeting data scientists and developers with limited cryptographic backgrounds, facilitating advanced data analysis in a secure environment. ### The main contributions of the paper are summarized as follows: 1. **Universal Applicability**: The linear kernel trick is introduced into the HE field. The proposed method is independent of the underlying HE scheme or library and can work in synergy with any underlying optimization technique to improve performance. 2. **Dimension - Independent Efficiency**: It shows the characteristic of approximately constant - time complexity, which is applicable to ML/STAT algorithms such as classification, clustering and dimension reduction, and especially has a significant reduction in execution time for high - dimensional data. 3. **Enhanced Training Potential**: It shows the potential to significantly improve ML training in the HE field, especially in cases where current HE training models are difficult to be competent. 4. **User - Friendly Method**: It is easy to be used by data scientists and developers with limited cryptographic knowledge. Through these improvements, the method proposed in the paper not only improves the efficiency of homomorphic encryption in high - dimensional data analysis, but also enables complex machine learning and statistical analysis to run efficiently in the encrypted domain.

Speed-up of Data Analysis with Kernel Trick in Encrypted Domain

Hyperdimensional Computing as a Rescue for Efficient Privacy-Preserving Machine Learning-as-a-Service

Phantom: A CUDA-Accelerated Word-Wise Homomorphic Encryption Library

Kernel homomorphic encryption protocol.

SHAPER: A General Architecture for Privacy-Preserving Primitives in Secure Machine Learning.

Toward Lossless Homomorphic Encryption for Scientific Computation

Faster CryptoNets: Leveraging Sparsity for Real-World Encrypted Inference

Design and Implementation of Big Data Analysis Algorithm in Ciphertext Domain Based on Homomorphic Encryption

HT2ML : An Efficient Hybrid Framework for Privacy-preserving Machine Learning Using HE and TEE

Towards Secure Big Data Analysis via Fully Homomorphic Encryption Algorithms

Towards realistic privacy-preserving deep learning over encrypted medical data

Logistic regression model training based on the approximate homomorphic encryption

Optimized Privacy-Preserving CNN Inference With Fully Homomorphic Encryption

Configurable Encryption and Decryption Architectures for CKKS-Based Homomorphic Encryption

Efficient CNN Building Blocks for Encrypted Data

MOFHEI: Model Optimizing Framework for Fast and Efficient Homomorphically Encrypted Neural Network Inference

Privacy preserving Neural Network Inference on Encrypted Data with GPUs

Efficient Encrypted Inference on Ensembles of Decision Trees

Privacy Preserving Inference for Deep Neural Networks: Optimizing Homomorphic Encryption for Efficient and Secure Classification

ESCAPED: Efficient Secure and Private Dot Product Framework for Kernel-based Machine Learning Algorithms with Applications in Healthcare

Privacy Preserving In-memory Computing Engine