Events 2021

📍 Satellite Event at JdS 2021 (Nice)

Launch of frENBIS

French local network of ENBIS – co-organised with the
Reliability and Uncertainty group of the SFdS

🧭 Practical Information

• 📅 Date: Friday, June 11, 2021
• ⏰ Time: 2:00 pm – 5:30 pm
• 📌 Location: Nice (Satellite event of JdS 2021)
• 🎯 Scientific theme:
  Stochastic approaches for the certification of machine learning algorithms

Organisation

• frENBIS:
  Yannig Goude, Bertrand Iooss, Jairo Cugliari, Anne Gégout-Petit, Jean-Michel Poggi
• Reliability and Uncertainty Group (SFdS):
  Chair: Mitra Fouladirad

🎯 Objectives of the Event

This half-day event marked the official launch of the
French ENBIS network (frENBIS), whose mission is to:

• promote the value of statistics for business and industry,
• foster interactions between industry and academia,
• strengthen links between the French statistical community and ENBIS.

The central topic concerned the certification of machine learning models for their integration into safety-critical systems.

The programme combined:

• industrial perspectives (aeronautics, railway, automotive),
• academic and methodological contributions,
• discussions on interpretability, regulation, and risk management, particularly in the energy sector.

A substantial amount of time was dedicated to open discussions, encouraging the emergence of shared challenges and opportunities for collaboration.

🎥 Access to Presentations

The presentations were recorded and are available to members via the
ENBIS Media Center.

🕒 Programme

Format: 25-minute talk + 15-minute Q&A

• 2:00 pm – Introduction (SFdS / frENBIS)
• 2:10 pm – Grégory Flandin
  (IRT Saint-Exupéry / DEEL Project)
  Machine Learning in Certified Systems
• 2:50 pm – Jayant Sen Gupta
  (Airbus AI Research)
  Robustness to the Training Distribution
• 3:30 pm – Break
• 3:50 pm – Joseph Mikael
  (EDF R&D)
  Reinforcement Learning in Risk & Asset Management
• 4:30 pm – Freddy Lecue
  (Chief AI Scientist @ Thales Canada, Research Associate @ INRIA)
  Domain Knowledge and Explainable Machine Learning
• 5:10 pm – Open discussions
• 5:40 pm – End

📝 Abstracts

Grégory Flandin

Machine Learning in Certified Systems

Machine learning techniques are increasingly used to automate complex tasks. However, their integration into systems subject to certification constraints introduces new safety challenges. This talk analyses the risks associated with machine learning and discusses technical and organisational approaches to address certification requirements.

Jayant Sen Gupta

Robustness to the Training Distribution

Certifying machine learning models for safety-critical applications requires robustness with respect to the training distribution. This presentation explains why uniform performance over the support of the input distribution is a key objective, and presents ongoing work at Airbus within the DEEL and ANITI projects.

Joseph Mikael

Reinforcement Learning in Risk & Asset Management

Reinforcement learning methods have shown strong potential for improving risk and asset management decisions. However, their operational deployment raises important methodological and regulatory questions. This talk reviews these challenges and discusses current approaches to building trust in reinforcement learning systems.

Freddy Lecue

Domain Knowledge and Explainable Machine Learning

One of the main obstacles to deploying machine learning in critical systems is the lack of explicit justification for model decisions. This presentation explores how domain knowledge and knowledge graphs can enhance interpretability, illustrating approaches that combine machine learning with symbolic reasoning to produce more human-understandable explanations.