Regulatory Acknowledgment of Individual Autonomy in European Digital Legislation: From Meta-Principle to Explicit Protection in the Data Act

Abstract

Federated machine learning is a decentralized/ distributive method to train machine learning models by keeping the datasets of the collaborators private and safe at their respective sites. The collaborators send their model updates to a central aggregator which merges the local models into a global model and then sends the model updates to all the contributing parties. Thus, only model updates are sent ex-situ while the data remains in-situ. This method of working assumes the aggregator to be safe from malicious gradient tampering, poisoning attacks, and the introduction of backdoors, which is not always the case. So, in this work, a Blockchain-based federated learning method is proposed for the secure aggregation of private data. In the proposed blockchain-based model, aggregation is carried out as the clients submit their respective local updates, resulting in accuracy comparable to the traditional centralized models. Moreover, for large datasets, the model provides lower latency, and thus, it can be a good solution for real-time and near-real-time applications.