PhD Students
ADRIAENS Florian (Starting date : March 2014)
Synchronization in networks of coupled nonlinear dynamical systems
Thesis advisors: Dirk Aeyels, Lode Wylleman
We are studying networks of impulsively coupled oscillators, mainly their synchronisation behaviour. Impulsively coupled networks of oscillators will possibly form clusters, and more general synchronise, through the process of absorption. The goal is to derive analytical conditions for certain network models, such that synchronised dynamic behaviour is guaranteed.
APERS Simon (Starting date : September 2014)
Engineering and analysis of quantum networks
Thesis advisor: Dirk Aeyels, Lode Willeman, and Alain Sarlette
We want to perform an in-depth analysis, from a robust dynamical systems point of view, of the quantum analogues of typical network operations and control. We start with quantum random walks, and would like to extend uncovered mechanisms at the interplay of quantum systems and classical networks to stabilize technologically useful quantum dynamics.
DE CUYPERE Eline (Starting date: September 2010)
Analysis of coupled queues
Thesis advisor : Dieter Fiems
A coupled queueing system consists of multiple queues, served by a single server such that in every queue a customer leaves upon service completion, and such that service is blocked if any of the queues is empty. Coupled queueing systems are frequently encountered in various contexts including assembly, production and communication networks. While conceptually simple, the analysis and performance of coupled queues is a challenging task, which draws upon different analysis methodology. Indeed, the state space of the associated Markov chains grows very fast in the number of queues involved and lead rapidly to a state space explosion. In this project, we aim to work on several numerical techniques such as the sparse-matrix techniques, the matrix-analytic methods and a novel numerical Taylor series expansions approach in order to cope with the state space explosion of the associated Markov chains.
DEKEMELE Kevin (Starting date : September 2015)
Thesis advisor: Mia Loccufier, Robin De Keyser
The research topic belongs to the discipline of mechanical vibration control and comprises passive and semi-active tuned mass dampers of the nonlinear type. Nonlinear spring-mass-damper elements are added locally to the main system. The structure, the number and the tuning parameters of the local additions have to be optimized to generate effective counteracting forces and as such mitigate excessive vibrations.
FARNAM Arash (Starting date : February 2014)
Distributed controllers inspired by PDEs and physical lattice models
Thesis advisor: Alain Sarlette
The goal is to design and analyze local control laws for distributed systems, by viewing them as an approximation of a PDE or lattice interaction where the control / design freedom is not at the boundary conditions but rather in the parameters and structure of the PDE, i.e. in the local coupling.
LOPATATZIDIS Stavros (Starting date : October 2012)
Robust Modeling and optimization in Stochastic Processes using Imprecise Probabilities with application to Queueing Theory
Thesis advisor : Gert de Cooman
MELANGE Willem (Starting date: September 2010)
Thesis advisor: Herwig Bruneel - Co-promotor: Joris Walraevens
This research is mainly focused on multiclass queueing models with global first-come-first-serve (gFCFS) service discipline, i.e., all arriving customers are accommodated in one single FCFS queue, regardless of their types. Queueing models are the de facto tools to model and analyse a broad range of systems where some kind of customers require some kind of service and whereby customers have to await their service in a queue. The main benefit of queueing models is that they include the stochastic nature of arrivals and service, which is present in many actual systems. Whereas traditional queueing research concentrates on servers that can serve all customers, this research considers multiple types of customers and dedicated servers, meaning that the server can only process customers of a specific type. As a consequence of the gFCFS rule, customers of one type may be blocked by customers of an other type, in that they may be unable to reach their dedicated server even at times when this server is idle, i.e., the system is basically non-workconserving. This blocking arises for instance in traffic.
VAN CAMP Arthur (Starting date : September 2011)
Thesis advisor : Gert de Cooman, co-promotor: Yves Van de Peer
Development of relevant theory and efficient algorithms for making inferences in credal threes, with applications in bio-informatics.
VERVISCH Bram (Starting date : September 2010)
Experimental modelling of rotating machinery for the design of prototypes
Thesis advisors : Mia Loccufier, Kurt Stockman
Rotating machinery can only operate properly if their vibration levels remain low at all times. Therefore no instabilities or sustained oscillations should occur. This research focuses on the identification, modelling and localisation of the various causes of energy dissipation. Especially rotating damping is one of the main topics because this type of damping can induce instability if the rotation speed too high.
ZHANG Zhifei (Starting date September 2013)
A constructieve approach to control with heterogeneity in distributed system networks
Thesis advisor : Alain Sarlette
This research focuses on methods to explicitly address heterogeneity in the control of distributed interacting networks. Catastrophic phenomena can happen when an algorithm designed for equal systems is applied to an even slightly heterogeneous network, see e.g. error patterns that destabilize distributed plants, or anomalies such as perpetual sliding in vehicle chains. It seems nontrivial to correct such behavior through robust control that would be compatible with the restrictions of distributed control. The goal is hence to develop such methods that incorporate robustness to specific structured heterogeneities in the coordination controller itself, as a complement to identification-based or adaptive control techniques. One concrete approach is e.g. for collective motion of robotica swarms, to build on nontrivial synchronization trajectories that would automatically average out the effects of deviating robot characteristics and/or measurement biases.
Post-Docs
CLAEYS Dieter
Analysis of queueing models with batch service
This research is mainly focused on queueing models with batch service. Queueing models are the de facto tools to model and analyse a broad range of systems where some kind of customers require some kind of service and whereby customers have to await their service in a queue. The main benefit of queueing models is that they include the stochastic nature of arrivals and service, which is present in many actual systems. Whereas traditional queueing research concentrates on servers that can serve one customer at a time, this research considers batch service, meaning that the server can process several customers simultaneously. Batch service arises for instance in transportation, manufacturing, telecommunications (packet aggregation), group screening, etc.
DE BOCK Jasper (Starting date: October 2015)
Developing the next generation of robust Bayesian networks: theory and efficient inference algorithms for mixed credal networks (FWO Postdoctoral fellowship)
Bayesian networks are powerful tools for modeling uncertainty, popular in fields such as machine learning and artificial intelligence. They can be used to describe complex problems in an intuitive way and provide tools for making decisions and answering questions related to these problems. Common fields of application are medicine, bioinformatics and engineering. However, the quantification of such a Bayesian network requires sharp, precise and trustworthy assessments of a large number of probabilities. If these are elicited from experts or learned from small data sets, this requirement is usually untenable and may render the resulting model and its inferences unreliable. This issue can be avoided by allowing for partial probability assessments such as bounds on probabilities, which can be assessed more reliably. The resulting generalized notion of a Bayesian network is called a credal network; its inferences are robust with respect to variations within the probability bounds that are given. However, this added feature of robustness comes at a price: the computational complexity of performing inference in credal networks tends to be very high. This project aims to introduce a new type of credal networks that does not suffer from this computational burden, by combining the best features of currently existing types of credal networks into a new theory. The main goal is to develop inference algorithms for this new type of credal networks that are as efficient as their Bayesian network counterparts.
QUAEGHEBEUR Erik (Starting date : October 2010)
Reasoning using desirability-based credal networks under epistemic irrelevante
Host : Gert de Cooman
In the field of artificial intelligence, probabilistic graphical models consisting of a network of connected nodes that represent the important aspects of a problem and their interactions have proven particularly useful. They allow us to describe complex problems in a manageable way and provide us with tools to answer questions and make decisions about issues related to these problems. Currently, most such models are limited by the fact that we must assume that we can assess the uncertainties present in the problem precisely or they do not allow us to model the interactions entirely realistically.
In this project, a mathematical theory will be formulated that describes a class of graphical models and the accompanying tools to deal with problems where the uncertainties cannot be assessed precisely and at the same time allow for a more realistic model of the interactions. These models result in answers with a precision that is commensurate with the information available about the uncertainties present and they constrain decisions in a way that is similarly appropriate. Based on the mathematical theory, algorithms for constructing and using them will be proposed, implemented in software, and applied to realistic test problems and a real life problem.
WYLLEMAN Lode
Synchronization of multi-agent networks