Federated Learning and Blockchain
To mitigate the inherent business risks when providing IoT data from different multiple stakeholders in complex farm-to-fork food supply chains, This project proposes an innovative system for enabling federated machine learning and artificial (edge) intelligence (ML/AI) supported with trust management platform based on private permissioned blockchain networks for managing data and analysis result transactions throughout supply chains and for external auditors (like regulatory bodies). It follows up on the EU H2020 PhasmaFOOD (www.phasmafood.eu) project where both partners FUB and VLF have already developed a ML/AI platform for food data analysis (mycotoxin detection, fish/meat spoilage, milk powder and oil adulteration) which is however centralized and has no privacy or trust attributes. By contrast to the centralized model, in the federated learning (FL) architecture each vertical IoT system will host and train a local model based on the stakeholder’s private data which are then aggregated trustfully and protected in a central model governed by the blockchain network. The main objectives are i) to jointly develop a decentralized federated learning and trust management blockchain network architecture and the corresponding ML/AI algorithms for supply chain and food analysis purposes, and ii) to implement a proof of concept platform (cloud, smartphone, IoT sensors) which demonstrates the full learning capabilities comparable to (centralized) PhasmaFOOD performance for the mentioned use cases but complies with privacy and trust attributes from the blockchain network. For leveraging and testing the platform the partners have access to food-related IoT data from the PhasmaFOOD project which can be used or emulated within multiple IoT vertical silos.