Compliance with safety standards is essential in safety-critical domains like automotive, rail and space. While traditional approaches to functional safety are well-established, the introduction of AI into safety-critical systems presents new complexities that challenge existing frameworks and methodologies due to the ‘black box’ characteristic of deep-learning models. Explainable AI (XAI) is vital for making AI decision-making processes transparent and understandable to human experts, and for ensuring safety and regulatory compliance.
Despite the value of XAI, there is currently a lack of systematic approaches to integration within AI-based systems and the Machine Learning lifecycle, especially in domains where safety is non-negotiable. This webinar seeks to address this gap by introducing the SAFEXPLAIN explainability by design approach.
This webinar will discuss the integration of XAI algorithms into the machine learning (ML) lifecycle for safety-critical systems. By embedding XAI principles from the outset, developers can create AI systems that are not only powerful but also transparent and compliant with functional safety (FUSA) standards.
The webinar will provide attendees with an overview of existing standards and research in XAI, introduce the SAFEXPLAIN approach, delve into the specific design of FUSA-aware DL solutions for the project’s use cases.
This is the second webinar in the SAFEXPLAIN webinar series. The first webinar, ‘Towards functional safety management for AI-based critical systems’, provided the framework and software architecture to incorporate AI-based solutions in safety-critical systems (top-down approach). This second seminar provides AI solutions, how to tailor them, and how to build the overall software architecture (bottom-up).
What you will learn
Webinar attendees will:
- Learn about the current challenges and gaps in integrating XAI with ML lifecycle processes.
- Explore a structured approaching to integrating XAI within the development and deployment of SAFEXPLAIN AI models to ensure compliance with functional safety standards.
- Gain insights into the innovative SAFEXPLAIN approach for leveraging AI in automotive, rail and space applications.
- Have access to the latest XAI research coming from the SAFEXPLAIN project (link to deliverable, website resources).
Speaker information
Robert Lowe is a researcher in artificial intelligence and driver monitoring Systems at the Research Institutes of Sweden (RISE AB). He is also an associate professor docent) in cognitive science at the University of Gothenburg, Department of Applied IT.
Robert joined RISE in 2023 as a senior researcher and is part of the Human-Centred AI Unit in the Department of Mobility and Systems. His research focuses on the application of AI in autonomous (and partially autonomous) systems, with a focus on research into using AI and models of driver behaviour for mitigating safety critical events as well as safe use of AI in such use cases. He researches, and coordinates, a number of projects related to the above within Sweden such as VInnova funded IntoxEye (REF: 2023-02606) and I-AIMS (REF: 2023-03068) and the SAFEXPLAIN project.