Abstract
The emerging topic of privacy-preserving deep learning as a service has attracted increasing attention in recent years, which focuses on building an efficient and practical neural network prediction framework to secure client and model-holder data privately on the cloud. In such a task, the time cost of performing the secure linear layers is expensive, where matrix multiplication is the atomic operation. Most existing mix-based solutions heavily emphasized employing BGV-based homomorphic encryption schemes to secure the linear layer on the CPU platform. However, they suffer an efficiency and energy loss when dealing with a larger-scale dataset, due to the complicated encoded methods and intractable ciphertext operations. To address it, we propose cuSCNN, a secure and efficient framework to perform the privacy prediction task of a convolutional neural network (CNN), which can flexibly perform on the GPU platform. Its main idea is 2-fold: (1) To avoid the trivia and complicated homomorphic matrix computations brought by BGV-based solutions, it adopts GSW-based homomorphic matrix encryption to efficiently enable the linear layers of CNN, which is a naive method to secure matrix computation operations. (2) To improve the computation efficiency on GPU, a hybrid optimization approach based on CUDA (Compute Unified Device Architecture) has been proposed to improve the parallelism level and memory access speed when performing the matrix multiplication on GPU. Extensive experiments are conducted on industrial datasets and have shown the superior performance of the proposed cuSCNN framework in terms of runtime and power consumption compared to the other frameworks.
Highlights
Deep learning (DL) has been applied to lots of fields [e.g., visual recognition (He et al, 2016), medical diagnosis (Shen et al, 2017), risk assessment (Deng et al, 2021a,b), and a recommender system (Shi et al, 2020; Wu et al, 2021a,b)], which achieves a superior performance in comparison with human cognition
The increasing popularity of cloud-based deep learning poses a natural question about privacy protection: if massive personal data are collected for model training and prediction, will this result in a rise in disclosing sensitive information? This paper focuses on tackling the privacy-preserving deep learning problem of a client that wishes to classify private images utilizing a convolution neural network (CNN) trained by a cloud server
We find that matrix-based computations are the core operations in the neural network prediction task
Summary
Deep learning (DL) has been applied to lots of fields [e.g., visual recognition (He et al, 2016), medical diagnosis (Shen et al, 2017), risk assessment (Deng et al, 2021a,b), and a recommender system (Shi et al, 2020; Wu et al, 2021a,b)], which achieves a superior performance in comparison with human cognition. For the proprietary of the model-holder, the DL model should not be revealed to users, in order to preserve their competitive advantage Following this mainstream, several solutions based on various secure computing technologies have been proposed, such as homomorphic encryption (HE)-based (Dowlin et al, 2016), multi-party computing (MPC)-based (Rouhani et al, 2018), and mixed-based solutions (Juvekar et al, 2018). The use of GPU technology to accelerate matrix multiplication is another important motivation of this study On this basis, we introduce cuSCNN, a practical realization of a mixed-based framework that supports the privacy-preserving prediction of convolutional neural networks (CNNs). We propose cuSCNN, an efficient and privacy-preserving neural network prediction framework that keeps user and server data secure.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
