Title: Master 2 Internship - Runtime Monitors of Neural Network in Aerospace Applications

Profile: Computer Science, AI

Level: Master 2 (MSc)

Duration: 5–6 months

Location: LAAS-CNRS, Toulouse, France (https://www.laas.fr)

Supervisors: Jérémie Guiochet (https://homepages.laas.fr/guiochet)

Salary: ~€600/month

Keywords: Computer Science, Artificial Intelligence, Machine Learning, Data Generation, Safety Monitoring

Technologies: Python, AI Frameworks

Artificial intelligence (AI) is increasingly being used in safety-critical applications, such as autonomous vehicles and surgical robotics. In these systems, perception tasks largely rely on deep neural networks (DNNs). However, the use of these components remains challenging because their behavior cannot be fully guaranteed, raising concerns about reliability and safety. Runtime monitoring mechanisms are among the promising approaches to address these limitations. Their purpose is to continuously monitor neural network behavior in order to detect or prevent potential errors. Two main families of monitors can be distinguished: rule-based monitors, which verify whether predefined safety properties are satisfied, and data-driven monitors, which are themselves developed through machine learning, often using neural networks. Evaluating the performance of these monitors is essential, particularly in situations where the main neural network produces incorrect predictions. This raises several important questions: What data should be used to evaluate these monitors? How should such data be collected, generated, and selected? How can the dataset cover both nominal and abnormal operating conditions? The objective of this internship is to investigate data generation and selection strategies for monitor evaluation and to develop a framework for testing and evaluating one or more monitoring mechanisms proposed by the AI research community. The internship will be conducted as part of an aviation-related case study, focusing on runway detection during aircraft landing approaches. It builds on the LARD (Landing Approach for Runway Detection) tool and dataset, which we have extended into Lardon-x, a tool currently under development at ANITI and LAAS-CNRS.

Application : CV + cover letter + transcripts of records for the last two academic years on the website