Privacy-Preserving Blockchain-Based Federated Learning for IoT
Devices (2021)
Objective
To propose a privacy-preserving blockchain-based federated learning system that
enables home appliance manufacturers to improve service quality and
functionality based on customer data while ensuring data privacy.
Methodology
Customers train the initial model provided by the manufacturer using both their
mobile phones and the mobile edge computing (MEC) server. Using their phones,
customers collect data from various home appliances, and then they download
and train the initial model with their local data. After deriving local models,
customers sign on to 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.
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
Strength
To protect customers’ privacy and improve test accuracy, enforced differential
privacy on the extracted features and proposed a new normalization technique.
Also, an incentive mechanism to award the participants is designed.
Use of Algorand consensus pro`tocol i.e accommodate more number of
participants.
Limitation
Delays in data models from crowdsourcing.
Potential challenges with heterogeneous storage and computation capabilities of
various IoT devices.
Application
Can be applied in smart home environments, where manufacturers use
aggregated customer data from home appliances to enhance product
development and service delivery.
Future Direction
Conduct more experiments and test the system with real-world home appliance
datasets. Moreover, we will strive to find the deterministically optimal balance
between local epochs and global epochs to obtain better test accuracy.
Federated Learning and Blockchain-Enabled Fog-IoT Platform for
Wearables in Predictive Healthcare (2023)
Objective
Proposed a fog-based IoT platform that leverages federated learning and private
blockchain technology to enhance data privacy and service integrity in predictive
healthcare environments utilizing wearable IoT devices.
Methodology
Establishes a distributed network combining local processing in wearable devices
and oversight via fog nodes. FL implements local model training on devices,
sending only model updates to fog nodes and preserving patient privacy. Utilizes
a private blockchain for secure communication and access control, creating an
immutable ledger for local modal training. Developed a low-cost solution using
Raspberry Pi devices to evaluate performance under real-world conditions.
Strength
It uses low-cost Raspberry Pi for implementation and demonstrates cost-
effectiveness in developing IoT solutions.
Incorporates federated learning and blockchain to enhance data privacy and
security.
Modular design allows for easier adaptability and integration of new wearable
devices.
Limitation
Variability in the processing capabilities of different wearable devices may lead
to inconsistent model performance.
Increasing numbers of connected devices could strain fog infrastructure, risking
performance degradation and response times.
Meeting strict healthcare regulations (e.g., HIPAA) adds complexity and legal
considerations to the platform's design and deployment because accuracy is
90%
Application
Focuses on predictive healthcare services utilizing wearable IoT devices.
Enables remote patient monitoring to improve healthcare delivery and efficiency.
Future Directions
Suggests exploring the integration of advanced machine learning techniques for
improved predictive analysis.
Adaptive Resource Allocation for Blockchain-Based Federated Learning
in the Internet of Things
Objective
The paper aims to propose a blockchain-based federated learning (FL) framework
for the Internet of Things (IoT) that addresses privacy concerns and
communication overhead while ensuring reliable model training among
distributed IoT devices. The framework seeks to enhance resource allocation
using Deep Reinforcement Learning (DRL)
Methodology
The proposed framework utilizes blockchain technology to facilitate
decentralized model training while preventing data leakage and enhancing
trustworthiness.
A dynamic mechanism of Committee-Based Participant selection is designed to
select participating nodes for model training, ensuring only reliable and high-
quality model updates are contributed.
The Resource Allocation problem is formulated as a Markov Decision Process
(MDP), and a DRL-based algorithm is implemented for adaptive channel
allocation, block size adjustment, and block producer selection to minimize
energy consumption while maintaining model convergence
Strengths
The framework eliminates the need for raw data sharing, mitigating privacy risks.
Introduces an accuracy-based incentive method to encourage effective
participation from IoT devices.
The use of DRL allows for real-time adaptation to changing environmental
conditions, enhancing the system’s efficiency.
Limitations
The implementation of DRL and blockchain may introduce computational
overhead that could affect scalability in resource-constrained environments.
Performance may vary under different data distributions (i.i.d. vs. non-i.i.d.),
potentially limiting effectiveness in all scenarios.
Practical Byzantine Fault Tolerance (PBFT) faces limitations imposed by network
bandwidth due to the frequent communication exchanges it requires
Application
The blockchain-based federated learning framework can be applied in remote
healthcare monitoring systems, allowing wearable devices to train AI models
collaboratively while preserving patient privacy. It is also useful in mobile health
applications, enabling smartphones to process local data efficiently without
overwhelming limited resources.
Future Direction
Further work is in progress to consider transfer learning, efficient consensus
algorithm, and important-based updating for implementing blockchain in FL-
enabled IoT scenarios
Secure and Efficient Federated Learning Through Layering and Sharding
Blockchain
Objective
The paper's primary objective is to design a secure and efficient federated
learning (FL) framework by implementing layering and sharding blockchain
technologies to compensate for the constraints imposed by blockchain
throughputs.
Methodology
Proposes ChainFL, a novel two-layer blockchain-driven FL system. It splits the IoT
network into multiple shards within the sub chain layer, reducing the scale of
information exchange, and employs a Direct Acyclic Graph (DAG)-based
mainchain as the mainchain layer, enabling parallel and asynchronous cross-
shard validation. Furthermore, the FL procedure is customized to integrate
deeply with the consortium blockchain, and a modified DAG consensus
mechanism is designed to mitigate distortion caused by abnormal models. To
provide a proof-of-concept implementation and evaluation, multiple sub chains
based on Hyperledger Fabric and a self-developed DAG-based mainchain are
deployed.
Strengths
ChainFL provides acceptable and sometimes better training efficiency (by up to
14%) and stronger robustness (by up to three times) compared to conventional
FL systems
Reduced the communication overhead and storage issues of Blockchain, using
sharding and IPFS respectively. Increased scalability. Test data was non-IID data
Limitations
Lack of an incentive mechanism for the devices.
Implementation was done on mini batch sizes only.
Applications
This federated learning framework has diverse applications in smart IoT
environments, such as intelligent transportation systems, smart grids, and
healthcare monitoring. By enabling collaborative learning while protecting
sensitive data, it can effectively drive advancements in personalized services
and smart industry solutions.
Future Work
plan to investigate model replacement or double-spending attacks in ChainFL
and explore an incentive mechanism based on blockchain technology to
encourage IoT device participation in FL tasks