A Blockchain-based Federated Learning Implementation Within a Recommendation System Landscape

In contrast with traditional centralized Artificial Intelligence techniques, Federated Learning promises to shift the paradigm by using decentralization to distribute training, acquainting not only user trust but also distinct contributions from multiple clients. In contrast, Blockchain became popular in recent years due to being a viable solution to deal with decentralized transactions in different contexts, guaranteeing accountability and integrity in the updates. This paper explores how these two methods cooperate specifically within a recommendation system scenario, where user personalization and data sensitivity are key concerns, through complete implementation, key benefits and challenges. After critically analyzing why blockchain may not be enough, an additional layer of cryptography is integrated. The results obtained are then presented in both scenarios, detailing how the model evolution is recorded on the blockchain, and how the model evolves with different numbers of valid clients that are able to train the model.