|
|
|
|
|
|
Schedule
Click on a link for more details
Show all the abstracts
Download the booklet of abtracts
Thursday 11:00 - 12:20 Routing Problems Room 138 - Chair: Pieter Vansteenwegen
Thursday 11:00 - 12:20 Emergency operations scheduling Room 130 - Chair: El-Houssaine Aghezzaf
Thursday 11:00 - 12:20 Algorithm design Room 126 - Chair: Gerrit Janssens
Thursday 11:00 - 12:20 Multiple Objectives Room 120 - Chair: Filip Van Utterbeeck
Thursday 13:30 - 14:50 Integrated logistics Room 138 - Chair: Kris Braekers
Thursday 13:30 - 14:50 Person transportation Room 130 - Chair: Célia Paquay
Thursday 13:30 - 14:50 Continuous models Room 126 - Chair: Nicolas Gillis
Thursday 13:30 - 14:50 Integer programming Room 120 - Chair: Bernard Fortz
Thursday 15:20 - 16:20 Material handling and warehousing 1 Room 138 - Chair: Greet Vanden Berghe
Thursday 15:20 - 16:20 Operations management Room 130 - Chair: Roel Leus
Thursday 15:20 - 16:20 Matrix factorization Room 126 - Chair: Pierre Kunsch
Thursday 16:30 - 17:10 Material handling and warehousing 2 Room 138 - Chair: Katrien Ramaekers
Thursday 16:30 - 17:10 Routing and local search Room 130 - Chair: An Caris
Thursday 16:30 - 17:10 Traffic management Room 126 - Chair: Joris Walraevens
Thursday 16:30 - 17:10 Pharmaceutical supply chains Room 120 - Chair: Bart Smeulders
Friday 10:50 - 12:10 Optimization in health care Room 138 - Chair: Jeroen Beliën
Friday 10:50 - 12:10 Network design Room 130 - Chair: Jean-Sébastien Tancrez
Friday 10:50 - 12:10 Local search methodology Room 126 - Chair: Patrick De Causmaecker
Friday 10:50 - 12:10 ORBEL Award Room 120 - Chair: Frits Spieksma
Friday 13:00 - 14:00 Production and inventory management Room 138 - Chair: Tony Wauters
Friday 13:00 - 14:00 Logistics 4.0 Room 130 - Chair: Thierry Pironet
Friday 13:00 - 14:00 Data clustering Room 126 - Chair: Yves De Smet
Friday 13:00 - 14:00 Collective decision making Room 120 - Chair: Bernard De Baets
Friday 14:10 - 15:10 Sport scheduling Room 138 - Chair: Dries Goossens
Friday 14:10 - 15:10 Discrete choice modeling Room 130 - Chair: Virginie Lurkin
Friday 14:10 - 15:10 Data classification Room 126 - Chair: Ashwin Ittoo
- A density-based decision tree for one-class classification
Sarah Itani (University of Mons) Co-authors: Fabian Lecron & Philippe Fortemps
- Comparison of active learning classification strategies
Xavier Siebert (Faculté Polytechnique de Mons) Co-authors: Nouara Bellahdid, Moncef Abbas Abstract: Modern technologies constantly produce huge quantities of data. Because these data are often plain and unlabeled, a particular class of machine learning algorithms is devoted to help in the data annotation process. In the setting considered in this paper, the algorithm interacts with an oracle (e.g., a domain expert) to label instances from an unlabeled data set. The goal of active learning is to reduce the labeling effort from the oracle while achieving a good classification. One way to achieve this is to carefully choose which unlabeled instance to provide to the oracle such that it most improves the classifier performance. Active learning therefore consists in finding the most informative and representative sample. Informativeness measures the impact in reducing the generalization error of the model, while representativeness considers how the sample represents the underlying distribution. In early active learning research the approaches were based on informativeness, with methods such as uncertainty sampling, or query by committee. These approaches thus ignore the distribution of the data. To overcome this issue, active learning algorithms that exploit the structure of the data have been proposed. Among them, approaches based on the representativeness criterion have proved quite successful, such as clustering methods and optimal experiment design. Various approaches combining the two criteria have been studied: methods based on the informativeness of uncertainty sampling or query by committee, and a measure of density to discover the representativeness criterion, others methods combine the informativeness with semi-supervised algorithms that provide the representativeness. In this work, we review several active learning classification strategies and illustrate them with simulations to provide a comparative study between these strategies.
- Risk bounds on statistical learning
Boris Ndjia Njike (Université de Mons) Co-authors: Xavier Siebert
|
|
|
ORBEL - Conference chair: Prof. A. Arda -
Platform: Prof. M. Schyns.
|
|
|
|