Privacy-Preserving Blockchain-Based Federated Learning for IoT Devices

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.

• The paper introduces a decentralized federated learning model utilizing blockchain to ensure data integrity and privacy in IoT smart home systems.
• The paper integrates differential privacy by adding noise to user data, thereby preventing reverse-engineering during model updates.
• It proposes an advanced normalization method that boosts model accuracy under privacy constraints and validates an incentive-based scheme to encourage honest participation.

Privacy-Preserving Blockchain-Based Federated Learning for IoT Devices

The paper "Privacy-Preserving Blockchain-Based Federated Learning for IoT Devices" addresses the integration of federated learning (FL) with blockchain technology to enhance the privacy and efficiency of internet of things (IoT) devices, particularly focusing on smart home systems. The authors propose a decentralized approach leveraging FL and blockchain to allow home appliance manufacturers to build machine learning models using data collected from IoT devices without compromising user privacy.

Key Contributions

1. Decentralized Federated Learning System: The paper introduces a federated learning architecture that employs blockchain technology to manage data and model updates in a decentralized manner. Unlike traditional centralized FL systems, this approach mitigates reliance on a single server, reducing risks associated with central data aggregation. The blockchain serves as a tamper-proof ledger that ensures transparency and accountability among participating entities.
2. Privacy Preservation through Differential Privacy: To protect user data from being reverse-engineered via model updates, the authors integrate differential privacy (DP) into the FL framework. They insert noise into the extracted features of data before uploading it to the blockchain, providing rigorous privacy guarantees.
3. Normalization Technique: The paper proposes an innovative normalization method that improves the test accuracy of the FL model under DP constraints. By relaxing constraints of batch normalization, the new technique optimizes the trade-off between privacy and model performance.
4. Incentive Mechanism: An incentive scheme is designed to attract users to participate in the FL tasks. The system uses a reputation-based model to reward honest participants and penalize malicious actors attempting to compromise the learning process through data poisoning attacks.

Methodology and Results

• System Design: The system employs a hierarchical model where smartphones collect data from home appliances and perform initial training. The blockchain records the updates and aids in aggregating these into a global model. Privacy is maintained by incorporating DP noise.
• Evaluation: Experiments utilizing the MNIST dataset demonstrate the robustness of the proposed system. The results suggest that under differential privacy, the proposed normalization technique outperforms traditional batch normalization in terms of test accuracy. The study also examines the computational feasibility of deploying federated learning on resource-constrained devices such as Raspberry Pi, indicating the method's practicality.
• Incentive Evaluation: The paper's approach to reputational incentives is validated, revealing its effectiveness in encouraging truthful model updates and deterring adversarial behavior.

Implications and Future Directions

The research significantly impacts the field of practical, privacy-preserving machine learning for IoT environments. By using blockchain to decentralize the FL process, it opens avenues for further research into secure, distributed learning systems. The privacy protection strategies discussed could be expanded to more sophisticated neural networks and other real-world datasets beyond smart home applications.

Future research could explore optimizing the number of global and local epochs to balance the trade-offs between privacy, computational cost, and model accuracy. The integration of more advanced incentive mechanisms and the exploration of other blockchain consensus algorithms might enhance system robustness and scalability, fostering broader applications in sectors requiring stringent privacy guarantees.