http://hdl.handle.net/2268/85 Search this channel search http://orbi.ulg.ac.be/simple-search http://hdl.handle.net/2268/148453 Title: Meta-learning of Exploration/Exploitation Strategies: The Multi-Armed Bandit Case <br/> <br/>Author, co-author: Maes, Francis; Wehenkel, Louis; Ernst, Damien <br/> <br/>Abstract: The exploration/exploitation (E/E) dilemma arises naturally in many subfields of Science. Multi-armed bandit problems formalize this dilemma in its canonical form. Most current research in this field focuses on generic solutions that can be applied to a wide range of problems. However, in practice, it is often the case that a form of prior information is available about the specific class of target problems. Prior knowledge is rarely used in current solutions due to the lack of a systematic approach to incorporate it into the E/E strategy. To address a specific class of E/E problems, we propose to proceed in three steps: (i) model prior knowledge in the form of a probability distribution over the target class of E/E problems; (ii) choose a large hypothesis space of candidate E/E strategies; and (iii), solve an optimization problem to find a candidate E/E strategy of maximal average performance over a sample of problems drawn from the prior distribution. We illustrate this meta-learning approach with two different hypothesis spaces: one where E/E strategies are numerically parameterized and another where E/E strategies are represented as small symbolic formulas. We propose appropriate optimization algorithms for both cases. Our experiments, with two-armed “Bernoulli” bandit problems and various playing budgets, show that the metalearnt E/E strategies outperform generic strategies of the literature (UCB1, UCB1-T UNED, UCB-V, KL-UCB and epsilon-GREEDY); they also evaluate the robustness of the learnt E/E strategies, by tests carried out on arms whose rewards follow a truncated Gaussian distribution. http://hdl.handle.net/2268/147763 Title: A First Measurement Look at the Deployment and Evolution of the Locator/ID Separation Protocol <br/> <br/>Author, co-author: Saucez, Damien; Iannone, Luigi; Donnet, Benoît <br/> <br/>Abstract: During the last decade, we have seen the rise of discussions regarding the emergence of a Future Internet. One of the proposed approaches leverages on the separation of the identifier and the locator roles of IP addresses, leading to the LISP (Locator/Identifier Separation Protocol) protocol, currently under development at the IETF (Internet Engineering Task Force). Up to now, researches made on LISP have been rather theoretical, i.e., based on simulations/emulations often using Internet traffic traces. There is no work in the literature attempting to assess the state of its deployment and how this has evolved in recent years. This paper aims at bridging this gap by presenting a first measurement study on the existing worldwide LISP network (lisp4.net). Early results indicate that there is a steady growth of the LISP network but also that network manageability might receive a higher priority than performance in a large scale deployment. http://hdl.handle.net/2268/145692 Title: Robust stress intensity factors evaluation for 3D cracks with X-FEM <br/> <br/>Author, co-author: Minnebo, Hans; Béchet, Eric; Moës, Nicolas http://hdl.handle.net/2268/145691 Title: Explicit dynamics with X-FEM to handle complex geometries <br/> <br/>Author, co-author: Moes, Nicolas; Rozycki, P.; Béchet, Eric http://hdl.handle.net/2268/145260 Title: Internet Topology Discovery <br/> <br/>Author, co-author: Donnet, Benoît <br/> <br/>Abstract: Since the nineties, the Internet has seen an impressive growth, in terms of users, intermediate systems (such as routers), autonomous systems, or applications. In parallel to this growth, the research community has been looking for obtaining and modeling the Internet topology, i.e., how the various elements of the network interconnect between themselves. An impressive amount of work has been done regarding how to collect data and how to analyse and model it. This chapter reviews main approaches for gathering Internet topology data. We first focus on hop limited probing, i.e., traceroute-like probing. We review large-scale tracerouting projects and discuss traceroute limitations and how they are mitigated by new techniques or extensions. Hop limited probing can reveal an IP interface vision of the Internet. We next focus on techniques for aggregating several IP interfaces of a given router into a single identifier. This leads to a router level vision of the topology. The aggregation can be done through a process called alias resolution. We also review a technique based on IGMP probing that silently collect all multicast interfaces of a router into a single probe. We next refine the router level topology by adding subnet information. We finish this chapter by discussing the AS level topology, in particular the relationships between ASes and the induced hierarchy. http://hdl.handle.net/2268/145217 Title: PRIM related correlations <br/> <br/>Author, co-author: Mathéi, Axel http://hdl.handle.net/2268/144646 Title: On the Number of Abelian Bordered Words <br/> <br/>Author, co-author: Rampersad, Narad; Rigo, Michel; Salimov, Pavel <br/> <br/>Abstract: In the literature, many bijections between (labeled) Motzkin paths and various other combinatorial objects are studied. We consider abelian (un)bordered words and show the connection with irreducible symmetric Motzkin paths and paths in Z not returning to the origin. This study can be extended to abelian unbordered words over an arbitrary alphabet and we derive expressions to compute the number of these words. In particular, over a 3-letter alphabet, the connection with paths in the triangular lattice is made. Finally, we study the lengths of the abelian unbordered factors occurring in the Thue--Morse word. http://hdl.handle.net/2268/144561 Title: Extremely Randomized Trees and Random Subwindows for Image Classification, Annotation, and Retrieval <br/> <br/>Author, co-author: Marée, Raphaël; Wehenkel, Louis; Geurts, Pierre <br/> <br/>Abstract: We present a unified framework involving the extraction of random subwindows within images and the induction of ensembles of extremely randomized trees. We discuss the specialization of this framework for solving several general problems in computer vision, ranging from image classification and segmentation to content-based image retrieval and interest point detection. The methods are illustrated on various applications and datasets from the biomedical domain http://hdl.handle.net/2268/143553 Title: Travaux Pratiques de Télédétection Spatiale <br/> <br/>Author, co-author: Denis, Antoine <br/> <br/>Abstract: Notes de travaux pratiques d'initiation à la télédétection spatiale. Le logiciel principalement utilisé dans ces travaux est ENVI de la firme Exelis. Les autres logiciels sont WINDIP, ViewSpecPro et Google Earth. Les exercices concernent, par ordre d'importance: la classification supervisée d'une image haute résolution (observation, correction géométrique, identification des zones d'entrainement et caractérisation statistique des échantillons, classification, validation), l'analyse diachronique de 2 images basses résolution, l'analyse temporelle d'une série d'images basses résolution (profil NDVI), création de néo-canaux (NDVI, Tasseled cap, ACP) la visualisation 3D, l'analyse de données (image et ponctuelle) hyperspectrales, la découverte de Google Earth, la recherche d'images satellites sur le web,... http://hdl.handle.net/2268/143530 Title: Optimal discovery with probabilistic expert advice: finite time analysis and macroscopic optimality <br/> <br/>Author, co-author: Bubeck, Sébastien; Ernst, Damien; Garivier, Aurélien <br/> <br/>Abstract: We consider an original problem that arises from the issue of security analysis of a power system and that we name optimal discovery with probabilistic expert advice. We address it with an algorithm based on the optimistic paradigm and on the Good-Turing missing mass estimator. We prove two different regret bounds on the performance of this algorithm under weak assumptions on the probabilistic experts. Under more restrictive hypotheses, we also prove a macroscopic optimality result, comparing the algorithm both with an oracle strategy and with uniform sampling. Finally, we provide numerical experiments illustrating these theoretical findings. http://hdl.handle.net/2268/143530 Title: Optimal discovery with probabilistic expert advice: finite time analysis and macroscopic optimality <br/> <br/>Author, co-author: Bubeck, Sébastien; Ernst, Damien; Garivier, Aurélien <br/> <br/>Abstract: We consider an original problem that arises from the issue of security analysis of a power system and that we name optimal discovery with probabilistic expert advice. We address it with an algorithm based on the optimistic paradigm and on the Good-Turing missing mass estimator. We prove two different regret bounds on the performance of this algorithm under weak assumptions on the probabilistic experts. Under more restrictive hypotheses, we also prove a macroscopic optimality result, comparing the algorithm both with an oracle strategy and with uniform sampling. Finally, we provide numerical experiments illustrating these theoretical findings. http://hdl.handle.net/2268/142763 Title: A Learning Procedure for Sampling Semantically Different Valid Expressions <br/> <br/>Author, co-author: Lupien St-Pierre, David; Maes, Francis; Ernst, Damien; Louveaux, Quentin http://hdl.handle.net/2268/142672 Title: Survival analysis: finding relevant epistatic SNP pairs using Model- Based Multifactor Dimensionality Reduction <br/> <br/>Author, co-author: Van Lishout, François; Vens, Céline; Urrea, Victor; Calle, M. Luz; Wehenkel, Louis; Van Steen, Kristel <br/> <br/>Abstract: Analyzing the combined effects of genes (and/or environmental factors) on the development of complex diseases is quite challenging, both from the statistical and computational perspective, even using a relatively small number of genetic and non-genetic exposures. Several data-mining methods have been proposed for interaction analysis, among them, the Multifactor Dimensionality Reduction Method (MDR). Model-Based Multifactor Dimensionality Reduction (MB-MDR), a relatively new dimensionality reduction technique, is able to unify the best of both nonparametric and parametric worlds, and has proven its utility in a variety of theoretical and practical settings. Until now, MB-MDR software has only accommodated traits that are measured on a binary or interval scale. Time-to-event data could therefore not be analyzed with the MB-MDR methodology. MB-MDR-3.0.0 overcomes this shortcoming of earlier versions. We show the added value of MB-MDR for censored traits by comparing the implemented strategies with more classical methods such as those based on a parametric regression paradigm. The simulation results are supplemented with an application to real-life data. http://hdl.handle.net/2268/142449 Title: Sparse Summarization of Robotic Grasp Data <br/> <br/>Author, co-author: Hjelm, Martin; Ek, Carl Henrik; Detry, Renaud; Kjellström, Hedvig; Kragic, Danica http://hdl.handle.net/2268/142448 Title: Learning a Dictionary of Prototypical Grasp-predicting Parts from Grasping Experience <br/> <br/>Author, co-author: Detry, Renaud; Ek, Carl Henrik; Madry, Marianna; Kragic, Danica http://hdl.handle.net/2268/142357 Title: Optimized Look-Ahead Trees: Extensions to Large and Continuous Action Spaces <br/> <br/>Author, co-author: Jung, Tobias; Ernst, Damien; Maes, Francis <br/> <br/>Abstract: This paper studies look-ahead tree based control policies from the viewpoint of online decision making with constraints on the computational budget allowed per decision (expressed as number of calls to the generative model). We consider optimized look-ahead tree (OLT) policies, a recently introduced family of hybrid techniques, which combine the advantages of look-ahead trees (high precision) with the advantages of direct policy search (low online cost) and which are specifically designed for limited online budgets. We present two extensions of the basic OLT algorithm that on the one side allow tackling deterministic optimal control problems with large and continuous action spaces and that on the other side can also help to further reduce the online complexity. http://hdl.handle.net/2268/142251 Title: Principles of QoS in group communications <br/> <br/>Author, co-author: Mathy, Laurent; Edwards, C.; Hutchison, D. http://hdl.handle.net/2268/142250 Title: Scalable adaptive hierarchical clustering <br/> <br/>Author, co-author: Mathy, Laurent; Canonico, R.; Simpson, S.; Hutchison, D. http://hdl.handle.net/2268/142249 Title: The internet: A global telecommunications solution? <br/> <br/>Author, co-author: Mathy, Laurent; Edwards, C.; Hutchison, D. http://hdl.handle.net/2268/142248 Title: Towards TCAM-based Scalable Virtual Routers <br/> <br/>Author, co-author: Luo, Layong; Xie, Gaogang; Uhlig, Steve; Mathy, Laurent; Salamatian, Kavé; Xie, Yingke