Enhancing federated learning through blockchain : addressing challenges and unlocking potential

Thumbnail Image
Ανδρονικίδης, Γεώργιος
Journal Title
Journal ISSN
Volume Title
The rapid advancements in data-driven decision-making have increased the need for effective machine learning and deep learning models, which often require vast amounts of data for training. A groundbreaking methodology called Federated Learning (FL) enables collaborative machine learning without the need for centralized training data. FL addresses various challenges identified in conventional machine learning including privacy, data ownership, and communication bottlenecks. FL enables training on decentralized edge devices while aggregating local models into a global model. However, FL has several drawbacks such as security and integrity, accountability and transparency, and model provenance and security. This thesis explores the potential of leveraging blockchain technology to address these challenges and enhance FL capabilities. Blockchain’s decentralized and immutable nature offers several advantages, such as transparency, tamper resistance, and trustworthiness, which are crucial for addressing those challenges. By utilizing blockchain, key challenges such as model provenance and security, data privacy, model poisoning, and others can be effectively addressed. The integration of blockchain technology into the FL process has the potential to revolutionize the field by enabling a safer, more accountable, transparent, and trustworthy model. Future research directions and potential implementation challenges are also highlighted to encourage further exploration and development in this emerging field.
Blockchain, Federated learning, Lightweight, Scalable, Privacy