Authors:
Afef Awadid
1
;
Boris Robert
2
and
Benoît Langlois
3
Affiliations:
1
Technological Research Institute SystemX, Palaiseau, France
;
2
Technological Research Institute Saint Exupéry, Toulouse, France
;
3
Thales, Vélizy-Villacoublay, France
Keyword(s):
Engineering Processes, Methodological Guidelines, Trustworthiness Environment, MBSE, Arcadia Method, AI-Based Critical Systems, Machine Learning.
Abstract:
Because of the multidisciplinary nature of the engineering of a critical system and the inherent uncertainties and risks involved by Artificial Intelligence (AI), the overall engineering lifecycle of an AI-based critical system requires the support of sound processes, methods, and tools. To tackle this issue, the Confiance.ai research program intends to provide a methodological end-to-end engineering approach and a set of relevant tools. Against this background, an MBSE approach is proposed to establish the methodological guidelines and to structure a tooled workbench consistently. In this approach, the system of interest is referred to as the "Trustworthiness Environment" (i.e. the Confiance.ai workbench). The approach is an adaptation of the Arcadia method and hence built around four perspectives: Operational Analysis (the engineering methods and processes: the operational need around the Trustworthiness Environment), System Analysis (the functions of the Trustworthiness Environmen
t), Logical Architecture and Physical Architecture (abstract and concrete resources of the Trustworthiness Environment). Given the current progress of the Confiance.ai program, this paper focuses particularly on the Operational Analysis, leading to the modeling of engineering activities and processes. The approach is illustrated with an example of a machine learning model robustness evaluation process.
(More)