TECNALIA (www.tecnalia.com) is a private, non-for-profit, applied research Centre of international excellence with a strong market orientation aiming at achieving the major impact in economic terms, through innovation and technological development. TECNALIA is the leading private and independent research and technology organization in Spain and one of the largest in Europe, employing around 1,381 people (272 PhDs) and with an income of 117 million € in 2021.
In TERMINET project, three different groups of the TECNALIA`s DIGITAL Unit are working together:
- Artificial Intelligence, which fields of expertise among others are: Machine learning, Advanced algorithmics and Smart optimisation.
- High-Performance Computing for Big Data, which fields of expertise among others are: Big Data specific Infrastructures, Hybrid Big Data models (real-time/batch), and Edge/Fog specific solutions.
- Cybersecurity and Trust which fields of expertise among others are: Security and Privacy by Design and Distributed Ledger Technologies
TECNALIA has two main contributions in TERMINET project. The first one is to lead the WP4, providing its experience in data analytics, Federated Machine Learning (ML), and Big Data (BD) technologies and the second one is to participate actively un WP5, providing its experience on Privacy Enhancing Technologies (PET) and Blockchain. TECNALIA contributes also to the definition of the requirements and the TERMINET architecture and participates as technology provider in UC2 and UC5.
Based on the combination of the experience in Data analytics, Federated Machine Learning, Big Data, Privacy Enhancing Technologies and Blockchain, TECNALIA is developing a Secure Federated Learning Framework (SFLF). This framework, explained in detail in the deliverable D4.2 “Distributed Intelligence Using AI and ML”, provides a decentralized approach for Federated Learning whose main particularity is that a central server in charge of performing the aggregation and training orchestration is not needed anymore, reducing Single Point of Failure issues as well as some confidentiality concerns. SFLF is composed mainly by i) a component for local training, ii) a component-client-blockchain in charge of sending and receiving notifications from the blockchain services; it is responsible for orchestrating and synchronizing all the process, iii) a Multi-Party Computation (MPC) component that is responsible for performing the aggregation process in a distributed and confidential way, and iv) a component for local inference.
This framework is going to be deployed in two of the six use cases of TERMINET project: Use Case 2: Pathway of Personalized Healthcare and Use Case 5: Group Training Surgery Using VR-enabled IoT Technologies as both require privacy and decentralization.
Summarizing, Secure Federated Learning Framework developed by TECNALIA presents an innovative approach removing the central server of the traditional FML architecture, incorporating Smart Contracts deployed over a Blockchain network to perform the aggregation function without a central service and using MPC technology to allow more secure computation functions.