Efficient Integer Vector Homomorphic Encryption Using Deep Learning for Neural Networks

Xie, Tianying; Li, Yantao

doi:10.1007/978-3-030-04167-0_8

Cited by 7 publications

(3 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In 2020, Li et al 12 introduced a third party to be the decryptor for decrypting the results from the cloud server and storing the results for request queries from the clients, which keeps the whole DNNs inference on the cloud server. Furthermore, in the previous work, Xie et al 13 proposed a method for processing encrypted data in DNNs inference with the same encryption scheme EIVHE 6 .…”

Section: Related Workmentioning

confidence: 99%

BatchEncryption: Localized Federated Learning in Preserving-privacy with Efficient Integer Vector Homomorphic Encryption

Xie

2023

Preprint

Self Cite

View full text Add to dashboard Cite

Deep Learning as a Service (DLaaS) provides an efficient way to facilitate deep neural networks (DNNs) in various applications. Despite the accuracy of deep learning models, there is a paramount need for robust protocols to protect the privacy and security of data, particularly sensitive information. Currently, It is not a good way to permit cloud models to process such data without proper safeguards, leading to potential privacy breaches. Especially in the training phase of the model, most models are trained by plaintext rather than ciphertext, which will pose potential privacy leakage of data owners. While some researchers have begun to use homomorphically encrypted data for training, most researchers only implement a single pair of keys for encryption and decryption, neglecting the importance of encryption scheme robustness when unique keys are employed by users. To address these issues, this paper presents a practical localized Federated Learning (FL) method named BatchEncryption using Efficient Integer Vector Homomorphic Encryption (EIVHE) for privacy-preserving training and inference phases. BatchEncryption encrypts raw datasets in blocks using different pairs of keys, which holds the diversity and robustness of the model. Our experiments demonstrate that deploying different pairs of keys to train neural networks holds higher accuracy than implementing a single pair of keys, and the proposed method achieves around 97% accuracy on the data encrypted by different pairs of keys on the MNIST dataset.

show abstract

Section: Related Workmentioning

confidence: 99%

BatchEncryption: Localized Federated Learning in Preserving-privacy with Efficient Integer Vector Homomorphic Encryption

Xie

2023

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…HE allows data to be processed in ciphertext, maintaining a trade-off between privacy and performance. To protect the user's privacy, the privacy-preserving NN is built to perform inference directly on encrypted data, the ciphertext inference can be done over two paradigms: interactive paradigm (online mode), such as GELU-NET [13] (model was partitioned NN into two non-colluding parties), GAZELLE [12] (high accuracy but huge memory cost), and BAYHENN [23] (inference was partitioned into linear and non-linear computations), and non-interactive paradigm (offline mode), such as CryptoNets [1] (the first work developed in this area), Chabanne et al [2] (first proposed work that combines activation function with BN function), Hesamiford et al [3] (suggested using derivative of polynomials to replace the activation function), Ishiyama et al [4] (proposed using CM for decreasing MD), Xie et al [24] (exploited the Efficient Integer Vector HE in CNN) and Dathathri et al [9] (initiated an optimizing compiler for HE NN inference). In this work, the focus is on ciphertext inference in the non-interactive paradigm.…”

Section: Privacy-preserving Deep Learningmentioning

confidence: 99%

CHE: Channel-Wise Homomorphic Encryption for Ciphertext Inference in Convolutional Neural Network

2022

Self Cite

View full text Add to dashboard Cite

Privacy-preserving deep learning (PPDL), which leverages the Homomorphic Encryption (HE), has attracted attention as a promising approach to ensure the privacy of deep learning applications' data. While recent studies have developed and evaluated the HE-based PPDL algorithms, the achieved performances, such as accuracy and latency, need improvement to make the applications practical. This work aims to improve the performance of the image classification of HE-based PPDL by combining two approaches -Channel-wise Homomorphic Encryption (CHE) and Batch Normalization (BN) with coefficient merging. Although these are commonly used schemes, their detailed algorithms and formulations have not been clearly described.The main contribution of current study is to provide complete and reproducible descriptions of these schemes. We evaluate our CHE and BN implementation by targeting the Cheon-Kim-Kim-Song scheme as an HE scheme and Convolution Neural Network (CNN) as a machine learning scheme while using the MNIST and CIFAR-10 as the datasets. In addition, we compare the results with the five state-of-the-art neural network architectures. Our experiments demonstrate that the CHE can serve as a tool for empirically achieving shorter latency (the shortest 7.76 seconds) and higher accuracy (the highest 99.32%) compared with the previous studies that aimed to establish the classification of the encrypted MNIST data with CNN. Our approach can aid in designing a more robust and flexible PPDL.

show abstract

“…Since our technique applies cryptography into significant learning, it might be acclimated to various diverse datasets before using them to continue. At the point when direct information encryption, others can't comprehend information without mystery key, and consequently it can secure protection and guarantee clients to utilize the neural system securely [18].In [19][20], we improve Gentry's holomorphic encryption to recommend an effective, disentangled mystery key holomorphic encryption. It in the picture encryption, with the goal that the picture can be prepared in encoded structure to secure the protection.…”

Section: Introductionmentioning

confidence: 99%

Accelerating Information Security in Cloud Computing using a Novel Holomorphic Scheme

V*,

2019

IJITEE

View full text Add to dashboard Cite

In the past decades, many security algorithms are invented and used in many applications for encryption and decryption of information or data. All the secure algorithms were uses the security key of size 8bit, 16bit, 32bit, 64bit and 128bits for encryption and decryption but literature survey says that higher the key size, higher the security. But the higher key size has number more bits and its requires more memory for storage and also to perform each and every bit through computational operations leads to more delay. To address these issues, the key and plain text are of 128bits and converted into integer. The converted integer values of key and plaintext are encrypted and decrypted using Holomorphic through Advanced Encryption Standard (AES). The research method is sophisticated and more protected through meaning fully lesser key in size and is accomplished for encryption in terms integer key and plaintext moderately than binary bits, therefore larger size of key and plaintext can be minimized and reduced the computational complexities. Finally the cipher text is uploaded into cloud through Real Time Transport Protocol (RTP) and Real Time Transport Control Protocol (RTCP) for storage. The main problem with continuous sharing of information into cloud is security attacks so in this research work, three different multimedia signals such as ECG, EEG and biomedical images are converted into integer and encrypted using AES. The stored data can access by the authorized users and can decode the information after decryption using AES.

show abstract

Efficient Integer Vector Homomorphic Encryption Using Deep Learning for Neural Networks

Cited by 7 publications

References 14 publications

BatchEncryption: Localized Federated Learning in Preserving-privacy with Efficient Integer Vector Homomorphic Encryption

BatchEncryption: Localized Federated Learning in Preserving-privacy with Efficient Integer Vector Homomorphic Encryption

CHE: Channel-Wise Homomorphic Encryption for Ciphertext Inference in Convolutional Neural Network

Accelerating Information Security in Cloud Computing using a Novel Holomorphic Scheme

Contact Info

Product

Resources

About