Revolution is messy. Presently and despite much activity, there is a significant lack of clarity regarding how to assure and approve machine learning (ML) for use in civil aircraft. This creates confusion, delay, and more importantly, barriers to innovation. This paper deobfuscates assurance and approval concepts for civil aircraft, and clarifies what these concepts mean for tackling the safe integration of ML into aircraft type designs. We overview applicable regulations and practices, with a focus on what they provide for, why, and how. We explain how the compliance context defines and organizes responsibility for substantiating that a system meets the safety intent of the regulations. We analyze concepts and structures central to assuring engineered systems and explain what they imply for ML-supported implementations. And we dispel some persistent misconceptions. Then we specify, locate, and bound the assurance burden for ML in accordance with the principles and structures identified. We call out and elaborate specifics that are needed, and name several that are not. As a result, an ill-defined and noisy conversation collapses into a smaller set of well-defined and actionable expectations for the assurance of ML-supported implementations in the civil aviation context.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Deobfuscating Machine Learning Assurance and Approval


    Beteiligte:


    Erscheinungsdatum :

    29.09.2024


    Format / Umfang :

    269359 byte





    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch





    Scenario-Based Methods for Machine Learning Assurance

    Hirschle, Manuel / Kirov, Dmitrii / Aievola, Rosario et al. | IEEE | 2023


    Toward Design Assurance of Machine-Learning Airborne Systems

    Dmitriev, Konstantin / Schumann, Johann / Holzapfel, Florian | TIBKAT | 2022


    Toward Design Assurance of Machine-Learning Airborne Systems

    Dmitriev, Konstantin / Schumann, Johann / Holzapfel, Florian | AIAA | 2022