9th International Conference on ICT for Sustainable Development

Authors - Jalpesh Vasa, Amit Thakkar, Dev Bhavsar, Pratham Patel
Abstract - Federated learning (FL) emerges as a revolutionary paradigm for collaborative machine learning, enabling training on decentralized, privacy-sensitive data. With its distributed nature, FL holds immense potential in various domains, encompassing healthcare, finance, and smart cities. However, this very decentralization introduces unique vulnerabilities susceptible to diverse attacks. This paper presents a comprehensive analysis of FL, delving into its architecture, applications, and potential attack vectors. This paper investigates a range of attacks, such as poisoning attacks, Sybil attacks, and reconstruction attacks, which pose threats to the integrity of models and the privacy of data within the context of Federated Learning (FL). We propose robust solutions to counter these threats, employing methodologies such as differential privacy, secure aggregation, and adversarial training. Through rigorous analysis and innovative solutions, our study aims to strengthen the implementation of FL while preserving its security and privacy aspects. Additionally, we showcase the diverse applications of FL across various sectors, demonstrating its potential for transformative impact. Ultimately, this paper contributes to a comprehensive understanding of FL, facilitating its secure and ethical advancement in today's data-driven world.