Stay informed on COVID-19 updates from EASARead more

Machine Learning Application Approval

MLEAP

The research objectives and expected outcome

The project deals with the approval of machine learning (ML) technology for systems intended for use in safety-related applications in all domains covered by the EASA Basic Regulation (Regulation (EU) 2018/1139).

Data-driven learning techniques are a major opportunity for the aviation industry but come also with a significant number of challenges with respect to the trustworthiness of ML and deep learning (DL) solutions.

EASA published its Artificial Intelligence Roadmap in February 2020, followed by a first major deliverable, a Concept Paper ‘First usable guidance for level 1 machine learning applications’ in April 2021. This concept paper lays down the basis of EASA future guidance for ML applications approval and identifies a number of areas in which further research is necessary to identify efficient and practicable means of compliance with the defined ‘AI trustworthiness’ objectives.

The intended short-term effect of this project will be to streamline the certification and approval processes by identifying concrete means of compliance with the learning assurance objectives of the EASA guidance for ML applications (levels 1, 2 and 3 as defined in the EASA AI Roadmap), with a specific focus on Level 1B and Level 2.

The achieved medium-term effect of the project will be to alleviate some remaining limitations on the acceptance of ML applications in safety-critical applications.

The requested output

The research results will be a set of reports identifying a set of methods and tools to address the following three important topics:

  • Guarantees on ‘machine learning model generalisation’
  • Guarantees on ‘Data completeness and representativeness’
  • Guarantees on algorithm and model robustness

Along with the project, at least one real-scale aviation use case should be developed to demonstrate the effectivity and usability of the proposed methods and tools. Those use cases should be developed in a software and hardware environment, accessible remotely by EASA or through software package deliveries to EASA. The essential life cycle artefacts developed in the project to address the different steps of the W-shaped process should be made available to EASA.

The work break-down structure of this project is the following:

  • Task 1: Methods and tools for the assessment of completeness and representativeness of data sets (training, validation, and test) in data-driven ML and DL
  • Task 2: Methods and tools for quantification of generalisation guarantees for ML and DL models
  • Task 3: Methods and tools for the verification of an ML algorithm and model robustness/stability
  • Task 4: Communication, dissemination, knowledge-sharing, stakeholder management
  • Task 5: Project management