• 9h45 :             Café d'accueil
• 10h00-11h00 : Richard Vinter, Imperial College London : Control of Lumped-Distributes Control Systems
• 11h00-11h30 : David Espes, UBO, Lab-STICC, CNRS UMR 6285 : Anomaly-based Intrusion Detection Systems
• 11h30-12h00 : Emmanuel Radoi, Zaynab Baydoun, UBO, Lab-STICC, CNRS UMR 6285 : AI-based People Detection in Indoor Environments

Abstracts

Control of Lumped-Distributed Control Systems (Richard Vinter):

Lumped-distributed control systems are collections of interacting sub-systems, some of which have finite dimensional state spaces (comprising `lumped' components) and some of which have infinite dimensional linear state spaces (comprising `distributed' components). Lumped-distributed control systems encountered, for example, in models of thermal or distributed mechanical devices under boundary control, when we take account of control actuator dynamics or certain kinds of dynamic loading effects.

This talk will focus on an important class of (possibly non-linear) lumped-distributed control systems, in which the control action directly affects only the lumped sub-systems and the output is a function of the lumped state variables alone. We give examples of such systems, including a temperature-controlled test bed for measuring semiconductor material properties under changing temperature conditions and robot arms with flexible links.

A key observation is that there exists an exact representation of the mapping from control inputs to outputs, in terms of a finite dimensional control system `with memory'. We use this representation to propose a computational approach to solving optimal control problems, in which the cost depends on the control and output variables. This offers advantages over traditional methods that involve the approximation of infinite dimensional state spaces by high dimensional linear subspaces.

Anomaly-based Intrusion Detection Systems (David Espes):

Anomaly-based detection in cybersecurity has become a major challenge for businesses. Although signature-based systems can detect the presence of known attacks, they cannot detect unknown (zero-day) attacks. For this, anomaly-based approaches are used, which learn the normal behavior of a system and detect an anomaly when the behavior differs too much from normal behavior. Semi-supervised or unsupervised artificial intelligence algorithms meet such needs. However, many parameters can influence the performance and prediction efficiency of these algorithms. First, feature selection approaches will be presented by comparing statistical methods with iterative approaches. Finally, semi-supervised detection approaches based on contrastive learning will be presented to detect behaviors that differ from the normal behavior of a system.

AI-based People Detection in Indoor Environments (Emanuel Radoi, Zaynab Baydoun):

In this presentation, we first focus on delay-Doppler analysis of IR-UWB signals reflected by moving people and objects in an indoor environment. The combination of their movement characteristics and the characteristics of the IR-UWB signal leads to some unexpected effects after Doppler focusing, such as spurious peaks and a shortened ambiguity window. We describe these new results revealed by our analysis and suggest a way to handle these aspects in the context of the considered application.

Secondly, we consider some machine/deep learning algorithms to resolve moving objects and people based on the corresponding delay-Doppler maps. Thus, we evaluate the performance of Support Vector Machines and three convolutional neural network architectures (AlexNet, GoogLeNet, and ResNet-18) across various signal-to-noise ratio levels. We also compare their effectiveness, assessed in terms of correct classification rate and confusion matrix.