VISION AND AMBITION IN SCIENCE-ENABLED TECHNOLOGY
Recent technological advancements are paving the way for a hyperconnected world driven by Artificial Intelligence (AI), particularly in the realm of data-driven societies. Interconnected devices facilitated by IoT and 5G/6G networks are set to create new data-driven scenarios, such as intelligent transport systems and autonomous vehicles, advanced healthcare services, and digitalized buildings. However, these developments also pose increased cybersecurity and privacy threats, underscoring the importance of robust digital ecosystems. Recognizing AI's pivotal role, the EU emphasizes the need for cybersecurity and data protection, especially concerning Machine Learning (ML) techniques' deployment. Centralized ML architectures raise significant privacy concerns, prompting the exploration of decentralized approaches like Federated Learning (FL). FL, while inherently privacy-preserving, presents challenges in maintaining the right balance between accuracy and privacy. REMINDER aims to explore privacy preservation approaches within FL, including anonymization, differential privacy, and advanced cryptography, to enhance security across distributed systems.
COMPREHENSIVE RESEARCH APPROACH
Innovative data-driven services rely on addressing security and privacy requirements throughout the data lifecycle. Current ML deployments, centralized in nature, raise privacy and performance concerns. Federated Learning (FL) offers a collaborative alternative but faces security challenges, particularly against adversarial attacks like data and model poisoning. Blockchain integration has been proposed to address coordination issues in FL but presents performance limitations. Differential Privacy (DP) is another approach to mitigate privacy concerns, especially in FL settings. Additionally, accounting for the dynamic behavior of distributed systems is crucial, considering factors like concept drift and data drifts. Ensuring data protection across storage, transfer, and processing remains complex, particularly in heterogeneous and dynamic environments with constrained devices. REMINDER aims to tackle these challenges through comprehensive research and innovative solutions in FL-enabled distributed systems, including applications in eHealth and smart buildings.
INTERDISCIPLINARY COLLABORATION
The REMINDER project exemplifies a high degree of interdisciplinarity through its consortium, combining expertise in cybersecurity, edge computing, and machine learning (ML), including federated learning (FL). The partnership involves UMU's strong background in these fields alongside UWE’s experience in ML applications for IoT and network security. AIT contributes with cryptographic methods to enhance data security and privacy. Together with industry partner SIE, the consortium will demonstrate the project's applicability in eHealth and smart buildings, showcasing how different scientific and technological domains can work together to solve complex problems in decentralized systems.
EVALUATING RISKS AND FLEXIBILITY
REMINDER faces challenges related to the computational demands of ML on constrained IoT devices. The project explores model pruning and compression techniques to reduce model size and computational needs, along with investigating lightweight ML models such as Random Neural Networks (RNNs) suitable for such environments. Additionally, it addresses risks associated with privacy-preserving techniques in FL, such as potential impacts on ML accuracy from methods like differential privacy. The project's approach remains flexible, adapting to technical findings and the specific needs of the planned use cases.
