Project Abstract

Federated learning (FL) has emerged as a distributed machine learning method for various Internet of Things (IoT) applications. Most of the existing IoT devices are resource-limited and unable to support federated learning that requires high processing capability, high power consumption, and large storage capacity.

Thus, heavy computation is offloaded to resource-powerful edge devices in order to mitigate the aforementioned challenges. In addition, federated learning is susceptible to security attacks, including global aggregation and distributed poisoning attacks. Thus, blockchain technology has been used to improve FL security and performance. However, none of the current research presents a blockchain-based FL suitable for resource-limited IoT devices.

The main research question is to explore and subsequently develop an efficient and secure public blockchain integration with FL for resource-limited devices. The proposed solution is expected to overcome the constraints limited by current approaches.

This solution will be designed using a series of programming languages, including but not limited to Solidity, Javascript, Python, and C++. The security and performance of the solution will be analyzed and evaluated with other related work. The security analysis includes analysis of smart contract vulnerabilities, private key storage, and some basic IoT security attacks, such as replay, tracking, and cloning attacks. The performance analysis of the proposed solution will be quantified in terms of computation cost, storage cost, and data persistence.