Multi-Dimensional Certification of Modern Distributed Systems

Abstract

Machine Learning (ML) is increasingly used to drive the operation of complex distributed systems deployed on the cloud-edge continuum enabled by 5G. Correspondingly, distributed systems' behavior is becoming more non-deterministic in nature. This evolution of distributed systems requires the definition of new assurance approaches for the verification of non-functional properties. Certification, the most popular assurance technique for system and software verification, is not immediately applicable to systems whose behavior is determined by Machine Learning-based inference. However, there is an increasing push from policy makers, regulators, and industrial stakeholders towards the definition of techniques for the certification of non-functional properties (e.g., fairness, robustness, privacy) of ML. This article analyzes the challenges and deficiencies of current certification schemes, discusses open research issues and proposes a first certification scheme for ML-based distributed systems.

Policy advice and best practices on bias and fairness in AI

“…A number of initiatives have started to standardize, audit, and certify algorithmic bias and fairness (Szczekocka et al, 2022), such as the IEEE P7003 TM Standard on Algorithmic Bias Considerations 7 , the IEEE Ethics Certification Program for Autonomous and Intelligent Systems 8 , the ISO/IEC TR 24027:2021-Bias in AI systems and AI aided decision making 9 , and the NIST AI Risk Management Framework 10 . Challenges of certification schemes are discussed in Anisetti et al (2023). Moreover, very few works attempt at investigating the practical applicability of fairness in AI (Madaio et al, 2022;Makhlouf et al, 2021b;Beutel et al, 2019), whilst several external audits of AI-based systems have been conducted (Koshiyama et al, 2021), sometimes with extensive media coverage (Camilleri et al, 2023).…”

Revisiting Trust Management in the Data Economy: A Road Map

“…They do not apply to data-intensive applications whose interactions are driven by (dynamic) non-functional requirements (e.g., [3]) referred also to data. In our reference example, trust must be negotiated on the basis of the geographical location of applications and data, to comply with data protection regulations (e.g., data of European citizens shall be processed in the EU or countries for which there exists an adequacy decision).…”

“…Service profiles should rely on upto-date information retrieved according to assurance techniques [5]. The integration with such techniques (e.g., certification [3]) must consider how to i) model (historical) service profiles, ii) compose and derive information from service profiles, iii) safely manage service profiles (C2.2, C2.3). The latter ensures that service profiles are stored and adequately protected for future negotiations, for instance, building on selective release [7], and certificate composition [6].…”

“…Within these environments, data consumers and providers collaborate to surmount existing legal and technical barriers to data sharing, thereby securely unlocking the full potential of their data. 3 The European Union (EU) is leading the way in this 2 https://joinup.ec.europa.eu/collection/elise-europeanlocation-interoperability-solutions-e-government/glossary/ term/data-marketplace 3 https://internationaldataspaces.org/, https://digitalstrategy.ec.europa.eu/en/library/staff-working-documentdata-spaces Month…”

QoS-Aware Deployment of Service Compositions in 5G-Empowered Edge-Cloud Continuum

“…In order to support QoS and constraintaware integration, the CSP exposes to the Deployment Agents suitable hooks to relevant resource (e.g., resource manager for deployment) or services (e.g., services offering security features) constituting their capabilities (R5). For instance, the CSP can offer a hook to access non-functional certificates (i.e., using certification scheme in [17], [18]) proving some capabilities or to invoke authentication services to support authentication requirements.…”