MSDP: multi-scheme privacy-preserving deep learning via differential privacy

Abstract

Human activity recognition (HAR) generates a massive amount of the dataset from the Internet of Things (IoT) devices, to enable multiple data providers to jointly produce predictive models for medical diagnosis. That the accuracy of the models is greatly improved when trained on a large number of datasets from these data providers on the untrusted cloud server is very significant and raises privacy concerns. With the migration of a deep neural network (DNN) in the learning experience in HAR, we present a privacy-preserving DNN model known as Multi-Scheme Differential Privacy (MSDP) depending on the fusion of Secure Multi-party Computation (SMC) and 𝜖-differential privacy, making it very practical since existing proposals are unable to make all the fully homomorphic encryption multi-key which is very impracticable. MSDP inputs a secure multi-party alternative to the ReLU function to reduce the communication and computational cost at a minimal level. With the aid of experimental verification on the four of the most widely used human activity recognition datasets, MSDP demonstrates superior performance with very good generalization performance and is proven to be secure as compared with existing ultramodern models without breach of privacy.

Temporal Differential Privacy for Human Activity Recognition

“…Leveraging these temporal sequences to train different models improves classification over the standard method. Moreover, the efficacy of this technique, employing varying window sizes for enhanced data representation, has also been validated in prior HAR studies [21], [27]. Hence in this paper, we propose TEMPDIFF (temporal differential privacy), an ensembling framework suited for time-series-based HAR that incorporates the temporality of the data to improve model training and provides differentially private predictions with competitive utility.…”

“…Recently differential privacy has been integrated into different types of HAR problems [10], [21], [33]. In [21], the authors present a privacy-preserving model based on secure multi-party computation that makes a fully homomorphic encryption multi-key. While this work is theoretically proved to be differentially private, it does not provide (ε, δ) results.…”

Privacy Preserving Image Retrieval Using Multi-Key Random Projection Encryption and Machine Learning Decryption

“…Following this track some researchers lean into not using DP as the sole privacy preserving technique, as it would lack the robustness of the needed privacy protection. Owusu et al, [34] has developed a privacy preserving deep neural network using secure MPC and DP fusion due to the impracticability of the FHE, their approach is secure with a good performance.…”

Privacy-Preservation Techniques for IoT Devices: A Systematic Mapping Study

“…Anonymization (31 papers) [34]- [64] Obfuscation (15 papers) [64]- [78] Multi-tier ML (5 papers) [79]- [83] Decentralized ML (10 papers) [78], [84]- [92] Cryptography (154 papers) [48], [51], [59], [62], [63], [76], [83], [85], [86], [92]- [236] Dataflow (51 papers) [61], [82], [83], [89], [101], [165], [216], [221], [223], [237]- [278] Data summarization (6 papers) [279]- [284] Personal data stores (2 papers) [285], [286] ensure privacy (a) blockchain which is used for verifiability and accountability of data collection, storage and access in IoT environments [31]; and (b) privacybased programming languages, which require information flows and privileges to be declared beforehand, so all the data elements are attached to respective policies [32]. 7) Data summarization: it is a process of creating a concise, yet informative, version of data to preserve the privacy of the original data.…”