[Paper Review] Privacy-Preserving Blockchain-Based Federated Learning for IoT Devices

Yang Zhao, Jun Zhao+6 more|arXiv (Cornell University)|Jun 26, 2019

Privacy-Preserving Technologies in Data64 references53 citations

TL;DR

Proposes a blockchain-based crowdsourcing federated learning framework for IoT, using differential privacy, IPFS off-chain storage, and a reputation-based incentive system, with an Algorand-based blockchain for auditability and security.

ABSTRACT

Home appliance manufacturers strive to obtain feedback from users to improve their products and services to build a smart home system. To help manufacturers develop a smart home system, we design a federated learning (FL) system leveraging the reputation mechanism to assist home appliance manufacturers to train a machine learning model based on customers' data. Then, manufacturers can predict customers' requirements and consumption behaviors in the future. The working flow of the system includes two stages: in the first stage, customers train the initial model provided by the manufacturer using both the mobile phone and the mobile edge computing (MEC) server. Customers collect data from various home appliances using phones, and then they download and train the initial model with their local data. After deriving local models, customers sign on their models and send them to the blockchain. In case customers or manufacturers are malicious, we use the blockchain to replace the centralized aggregator in the traditional FL system. Since records on the blockchain are untampered, malicious customers or manufacturers' activities are traceable. In the second stage, manufacturers select customers or organizations as miners for calculating the averaged model using received models from customers. By the end of the crowdsourcing task, one of the miners, who is selected as the temporary leader, uploads the model to the blockchain. To protect customers' privacy and improve the test accuracy, we enforce differential privacy on the extracted features and propose a new normalization technique. We experimentally demonstrate that our normalization technique outperforms batch normalization when features are under differential privacy protection. In addition, to attract more customers to participate in the crowdsourcing FL task, we design an incentive mechanism to award participants.

Motivation & Objective

• Design a hierarchical crowdsourcing FL system to help home appliance manufacturers improve services and predict consumer behavior.
• Provide privacy protections for participants by applying differential privacy to extracted features and a new normalization technique.
• Leverage a blockchain to audit model updates, prevent tampering, and enable traceability of contributions.
• Incorporate an incentive mechanism based on reputation and Multi-KRUM to reward reliable participants and deter poisoning.

Proposed method

• Model training is split into mobile feature extraction with epsilon-DP noise, followed by training fully connected layers on an MEC server.
• Use of IPFS for off-chain storage with model hashes recorded on the blockchain to handle large model sizes.
• A consortium blockchain with Algorand PoS/BFT consensus manages model submissions, verification with Multi-KRUM, and leader election for global model aggregation.
• A reputation-based incentive mechanism combines Multi-KRUM with VRF-selected verifiers to compute participant reputations and rewards or penalties.
• A novel normalization technique relaxes mean/variance constraints while bounding feature values to improve DP-perturbed training accuracy.

Experimental results

Research questions

• RQ1 How can blockchain and differentially private FL be integrated to securely aggregate models from distributed IoT participants?
• RQ2 Does the proposed normalization technique improve test accuracy under differential privacy compared to batch normalization?
• RQ3 Can a reputation-based incentive mechanism mitigate poisoning and encourage participant engagement in crowdsourced FL?
• RQ4 What storage and consensus strategies effectively handle large model updates in an IoT-centric FL setting?

Key findings

• Differential privacy is applied to features to protect participant data while enabling effective global model training.
• The proposed normalization technique yields higher test accuracy under differential privacy than batch normalization in experiments.
• Blockchain-based auditing via IPFS storage and Algorand consensus prevents tampering and enables traceability of model updates.
• An incentive mechanism leveraging reputation and Multi-KRUM helps identify reliable updates and deter malicious contributions.