Artificial Intelligence introduces a new paradigm to software development. While traditional algorithms are ex-plicitly programmed using instructions, in the machine learning approach the software is trained while exposed to data. The assurance of the result becomes directly dependent on the quality of the data. Data needs to be collected, processed, analyzed, validated, stored and protected. This study aims to establish the architecture for a framework where data management activities can be performed easily and efficiently. Regulation and standards about AI development for safety-critical applications are under development, but initial guidance already suggests a set of objectives that need to be satisfied during data management. A new concept of “learning assurance” is proposed to provide the adequate level of confidence. The concept proposed for data management can be applied with benefits in the quality of any safety-critical application based on machine learning, doesn't matter if it's in the aeronautical domain or others. The main contribution of this article, in addition to discussing the standards that will be applied in the aviation domain, is to bring in a practical way the tools that will assist in the development of machine learning systems and some ways to demonstrate compliance with the objectives of the standards in accordance with the roadmap of EASA for level 1 and level 2 machine learning safety-critical applications.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Framework Architecture for AI/ML Data Management for Safety-Critical Applications




    Publication date :

    2024-09-29


    Size :

    426540 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Monoprocessed computer architecture for safety critical applications

    de Almeida, J. R. / Fonseca, J. A. | British Library Conference Proceedings | 2002


    A Modeling Framework for Efficient Safety Critical Time-Triggered Architecture Design

    Cerrolaza, J. M.P. / Perez, A. / Berreteaga, O. et al. | British Library Conference Proceedings | 2007


    A modeling framework for efficient safety critical time-triggered architecture design

    Cerrolaza,J.M. / Perez,A. / Berreteaga,O. et al. | Automotive engineering | 2007


    A Modeling Framework for Efficient Safety Critical Time-Triggered Architecture Design

    de Olano, Alberto Ruiz / Perez, Antonio / Perez Cerrolaza, Juan M. et al. | SAE Technical Papers | 2007


    DACAPO: A Distributed Computer Architecture for Safety-Critical Control Applications

    Rostamzadeh, B. / Lonn, H. / Snedsbol, R. et al. | British Library Conference Proceedings | 1995