Abstract:The widespread integration of the Internet of Things (IoT) is crucial in advancing sustainable development. IoT service providers actively collect user data for analysis using sophisticated Deep Learning (DL) algorithms. This enables the extraction of valuable insights for business intelligence and improving service quality. However, as these datasets contain sensitive personal information, there is a risk of privacy breaches when DL models are employed. This vulnerability may result in Membership Inference Attacks (MIA), potentially leading to the unauthorized disclosure of highly sensitive data. Therefore, developing an efficient and privacy-preserving data analysis system for IoT is imperative. Recent research has highlighted the effectiveness of utilizing Bloom Filter (BF)-encoding in conjunction with Differential Privacy (DP) for safeguarding privacy during data analysis. Given its attributes of low complexity and high utility, this approach proves effective, particularly in resource-constrained IoT domains. With this in mind, we propose a novel framework for privacy-preserving IoT data analysis based on BF-encoded data. Our research introduces an innovative BF-encoding technique combined with Local Differential Privacy (LDP), capable of efficiently encoding various types of IoT data (such as facial images and smart-meter data) while maintaining privacy when integrated into DL algorithms for downstream analysis. Experimental results demonstrate that our BF-encoded data surpasses the utility of standard BF-encoded data when utilized in DL algorithms for downstream tasks, showcasing an approximate 30% improvement in classification accuracy. Furthermore, we assess the privacy of these DL models against MIA, revealing that attackers can only make random guesses with an accuracy of approximately 50%.

Bounding the Invertibility of Privacy-preserving Instance Encoding using Fisher Information

A Privacy Framework: Indistinguishable Privacy

A Privacy Framework

Is Private Learning Possible with Instance Encoding?

Fisher information under local differential privacy

Privacy Protectability: An Information-theoretical Approach

Inference Privacy: Properties and Mechanisms

Ensuring Privacy with Constrained Additive Noise by Minimizing Fisher Information

Inf2Guard: An Information-Theoretic Framework for Learning Privacy-Preserving Representations against Inference Attacks

Privacy for Fairness: Information Obfuscation for Fair Representation Learning with Local Differential Privacy

Privacy Preserving Calculation of Fisher Criterion Score for Informative Gene Selection

Privacy-Preserving Language Model Inference with Instance Obfuscation

FinPrivacy: A Privacy-preserving Mechanism for Fingerprint Identification

Privacy-Preserving Probabilistic Data Encoding for IoT Data Analysis

On the Gini-impurity Preservation for Privacy Random Forests

Inferentially-Private Private Information

Immersion and Invariance-based Coding for Privacy-Preserving Federated Learning

Mutual Information Bounded by Fisher Information

Private Inference in Quantized Models

FIBNet: Privacy-Enhancing Approach for Face Biometrics Based on the Information Bottleneck Principle

Towards Information Privacy for the Internet of Things