Training Models in a Privacy-preserving and Decentralized Fashion

The increasing volume of cyberattacks, the growing sophistication of malicious processes by cyber attackers, and the decentralization of interconnected devices have made cybersecurity a critical concern for companies and institutions. This has prompted them to develop more secure strategies to protect their data and systems. The consequences of a successful cyberattack can be devastating, resulting in financial loss, reputational damage, and even legal repercussions.

Artificial Intelligence (AI) is no longer just a buzzword. It’s becoming increasingly prevalent in our daily lives and has become a key focus in the field of computer science. The goal of AI is to create systems that can perform tasks that were once only possible for humans to do by simulating human cognitive abilities such as learning, reasoning, and self-correction.  

 One of the latest generations of AI to emerge is Decentralized Federated Learning (DFL), a novel federated system in which communications are decentralized among participants in a network. This means there is no central server to rely on, acting as a bottleneck or single point of failure, as is often the case in more traditional AI architectures. Network participants train Machine Learning (ML) models to solve specific tasks collaboratively. These participants do not exchange data in the clear but exchange parameters of the local models, increasing the privacy of the information. These participants could be mobile devices, computers, drones, or even organizations or companies. 

This new approach can be used in multiple application scenarios such as medical, Industry 4.0, mobile services, or military. 

• In medical scenarios, patient data can be enhanced by storing them in a secure and decentralized way or by tracking and analyzing patient data to prevent and improve medical care. In the same way, the collaboration between healthcare entities is ensured to create collaborative intelligent models for disease detection or vaccine discovery. 
• In Industry 4.0, decentralized collaborative models enable interoperability between machines to optimize processes, improve supply chain management, ensure process reliability and reduce costs. 
• In the military sector, it can be used to improve information security to prevent information leakage and reduce the risk of sabotage. It also ensures communication between devices on the battlefield and the delivery of critical information. 

Enrique Tomás Martínez Beltrán is working on his PhD thesis as a researcher for the Séneca Foundation in the Intelligent Systems and Telematics group at the University of Murcia. The main objective of his doctoral thesis is to develop a tool that enables various organizations to create collaborative AI models for different application scenarios. This tool will not only provide the capability of creating AI models using DFL but also measure the confidence of its predictions. It will be designed to be agnostic to user-defined data, models, or collaborative tasks and adaptable to different situations. The tool will feature a web interface, allowing users to interact with it and customize models. 

 This solution could help organizations develop more accurate and reliable AI models to enhance decision-making capabilities. This represents a significant technological advancement, allowing machines to learn collaboratively. The performance of the tool will be evaluated and compared to traditional systems and other similar projects, and it is expected to surpass them in terms of efficiency and robustness. 

Technological dependence, globalization, and ease of access to technologies make it possible for cyberattacks to materialize. The threat to information technology has never been more significant, and users need and demand cybersecurity in their environment. Even more so when terrorists can produce the attacks, criminal organizations, fanatical religious movements, intelligence services, or opposing military forces, in this context, our proposal would improve security features in cyberspace, avoiding attacks that affect deployed systems or devices.  

Among the most relevant collaborations in work is the “Cyber-Defence Campus at armassuise (Office fédéral de l’armement) Science & Technology” in the proposed DEFENDIS project, which aims to build a decentralized and federated learning framework for IoT device identification and security. The Swiss Army is interested in its direct applicability in various application scenarios, including improved decision-making and increased accuracy and speed of information processing. Some of the most notable use cases are detecting intelligent cyberattacks, communications between soldiers or drones on the battlefield, or detecting fake news in a decentralized way. Additionally, the University of Zurich (UZH) has also expressed interest in the wide range of services and applications that DEFENDIS opens up for the future. The outcomes of the DEFENDIS project will also be directly applicable to the EU-GUARDIAN project, in which we collaborate with several institutions and companies such as Indra Sistemas, Airbus Cybersecurity and the Bulgarian Defence Institute (BDI) among others. This recent project aims at creating a cutting-edge, accurate and reliable AI-based solution that operates and automates larger parts of incident management and cyber defence processes.

La Verdad. January 21, 2023. https://www.laverdad.es/ababol/nueva-forma-entrenar-20230121001054-ntvo.html 

 Enrique Tomás Martínez Beltrán | Ph.D. student in Computer Science at the University of Murcia