An approach based on NSGA-III algorithm for solving the multi-objective federated learning optimization problem

As the sheer volume of data continues to surge across both private and public networks, the potential for leveraging Machine Learning (ML) to streamline complex supply chains becomes increasingly viable. However, conventional centralized ML and data analytics systems are fraught with vulnerabilities, such as susceptibility to data breaches, loss of system control, and other malicious attacks. Enter Federated Learning (FL), a promising paradigm within ML that prioritizes data confidentiality and decentralized model training. Nevertheless, to harness its full potential and ensure optimal performance, an optimization process is required. This optimization challenge is further compounded by the fact that it falls into the category of NP-hard scheduling problems. Moreover, recent novel attacks on FL architecture have raised substantial security concerns. In this paper, we present a novel approach that marries Blockchain technology with the third-generation Non-dominated Sorting Genetic Algorithm (NSGA-III) to fortify the defense against attacks targeting FL algorithms operating within Internet of Things (IoT) systems. Our holistic optimization approach, underpinned by Blockchain, guarantees not only the veracity of trained models but also optimizes processing time and reduces the success rate of potential attacks. Consequently, we propose a hybrid methodology that integrates Blockchain and FL optimization, safeguarding ML models from a spectrum of threats while preserving user privacy. To tackle the bi-objective FL optimization problem, we introduce an efficient NSGA-III algorithm. Our rigorous experiments demonstrate the superior efficacy of our proposed solution compared to the current state-of-the-art approaches.