20th European Dependable Computing Conference
8-11 April 2025
Lisbon, Portugal

Abstract: There is a strong momentum towards data-driven services at all layers of society and industry. This started from large scale web-based applications such as Web search engines (e.g., Google, Bing), social networks (e.g., Facebook, TikTok, Twitter, Instagram) and recommender systems (e.g., Amazon, Netflix) and is becoming increasingly pervasive thanks to the adoption of handheld devices and the advent of the Internet of Things. Recent initiatives such as Web 3.0 are coming with the promise of decentralising such services for empowering users with the ability to gain back control over their personal data, and prevent a few economic actors from over concentrating decision power. However, decentralising online services calls for decentralising the data and the machine learning algorithms on which they heavily rely. While Federated Learning allows training machine learning models over decentralised data, it still relies on the centralised computation of model aggregations. In this presentation, I will present recent research works targeting the decentralisation of machine learning beyond the well know Federated Learning concept. A particular focus will be given on recent advances and open challenges for enforcing dependability and security in decentralized machine learning.

Abstract: Safety-critical systems (SCS) like those included in airplanes, cars, medical devices and nuclear plants go through a stringent domain-specific certification process to validate functional safety properties. The goal of such thorough verification and validation is to reduce the risk of operational failures leading to catastrophic consequences. Recently, there has been a surge of new SCS fueled by artificial intelligence (AI) including fully autonomous driving systems, AI-enabled medical devices, and robotics operators. Safety-related functionalities governed by AI tremendously escalate the complexity of achieving functional safety. The SCS industry is struggling with the necessary shift from low complexity hardware governed applications to these increasingly complex software-controlled systems executed on top of very complex processors. This comes at a time when society is starting to blindly (and dangerously) rely on such complex computing systems. There is thus an urgent need to find new approaches to manage the exponentially growing complexity of SCS in the era of AI. This talk reviews the main challenges associated with meeting functional safety requirements for safety-relevant AI functionalities and introduces a potential approach towards that end.

Abstract: The domains of railway and space systems both demand high levels of dependability and safety, yet they face unique challenges and adhere to different standards. This talk explores the application of dependability and safety principles in these two critical sectors, highlighting their specific challenges, commonalities, and differences. In the space domain, challenges include ensuring autonomy, safe communications, reconfigurations, long operational lifetimes, and the continuous availability of patch and dump functions. Additional concerns involve maintaining satellite battery charge, preventing system hangs or crashes, managing limited memory and processor resources, and detecting and correcting memory errors caused by single event upsets. Conversely, the railway domain focuses on passenger safety through rigorous and comprehensive hazard analysis, ensuring the operational readiness of emergency brakes, preventing unintended movements, controlling speed and signaling, and managing door operations, especially during transit. Both domains share commonalities such as the need for robust safety and dependability frameworks, though they follow different standards: ECSS and NASA standards for space, and CENELEC standards for railways. A key difference lies in certification requirements, with railways requiring certification and space systems not. Additionally, railways emphasize quantitative safety risk and hazard analysis, while space systems rely more on qualitative criticality analysis. This talk aims to provide insights into the distinct and overlapping aspects of dependability and safety in railway and space systems, offering a comprehensive understanding of their respective methodologies and practices.