|
|
|
|
|
|
Schedule
Thursday, 30 January |
08:30-09:10 | Welcome-registration-coffee | B14-B15-B16 |
09:10-09:30 | Opening session | Amphi Cuccaroni |
09:30-10:30 | Plenary session - Adam Letchford (Chair: Martine Labbé) | Amphi Cuccaroni |
10:30-11:20 | Coffee break | B14-B15-B16 |
11:20-12:40 | Parallel sessions |
|
Multi-objective Optimization Chair: Sara Tari Room: E501 |
Health-care Chair: Véronique François Room: E502 |
Public Transportation Chair: Javier Duran Micco Room: E503 |
Global Optimization Chair: Olivier Rigal Room: E601 |
Analytics 1 Chair: Rafael Van Belle Room: E602 |
12:40-14:10 | Lunch/ORBEL Board meeting | B14-B15-B16/E501 |
14:10-15:10 | Parallel sessions |
|
Air, Rail and Multimodal Transportation Chair: Paola Paregrini Room: E501 |
Game Theory Chair: Lotte Verdnock Room: E502 |
Transportation of People Chair: Yves Molenbruch Room: E503 |
Multi-level Optimization Chair: Concepcion Dominguez Room: E601 |
Analytics 2 Chair: Vedavyas Etikala Room: E602 |
15:10-15:50 | Coffee break | B14-B15-B16 |
15:50-16:40 | Parallel sessions |
|
Sport Timetabling Chair: Dries Goossens Room: E501 |
Project Management and Scheduling Chair: Fan Yang Room: E502 |
Rich Routing and Graphs Chair: Alexandre Bontems Room: E503 |
Logistics Chair: Yuan Yuan Room: E601 |
Analytics 3 Chair: Jari Peeperkorn Room: E602 |
16:50-17:30 | ORBEL general assembly | Amphi Cuccaroni |
19:00-20:00 | Cocktail | Bar Zango |
20:30-23:00 | Dinner | La Terrasse des Ramparts |
Friday, 31 January |
09:30-10:30 | Plenary session - Miguel F. Anjos (Chair: Luce Brotcorne) | Amphi Cuccaroni |
10:30-11:20 | Coffee break | B14-B15-B16 |
11:20-12:40 | Parallel sessions |
|
Automatic Configuration and Metaheuristics Analysis Chair: Kenneth Sorensen Room: E501 |
Real Life and Integrated Problems Chair: Jeroen Belien Room: E502 |
Warehouse Management Chair: Harol Mauricio Gamez Room: E601 |
Analytics 4 Chair: Noureddine Kouaissah Room: E602 |
12:40-14:10 | Lunch | B14-B15-B16 |
14:10-15:10 | Parallel sessions |
|
Orbel Award Chair: Jeroen Belien Room: E501 |
Optimization Chair: Julien Dewez Room: E502 |
Trucking Optimization Chair: Hatice Çalik Room: E601 |
Analytics 5 Chair: Diego Olaya Room: E602 |
15:10-15:50 | Coffee break | B14-B15-B16 |
15:50-16:50 | Parallel sessions |
|
Learning and Optimization Chair: Johan Van Kerckhoven Room: E501 |
Lot-sizing and Inventory Chair: Philippe Chevalier Room: E502 |
Rich Routing Chair: Cristian Aguayo Room: E601 |
17:00-17:15 | ORBEL award and closing session | Amphi Cuccaroni |
17:15-18:30 | Closing cocktail | B14-B15-B16 |
|
Thursday 11:20 - 12:40 Multi-objective Optimization Room E501 - Chair: Sara Tari
Thursday 11:20 - 12:40 Health-care Room E502 - Chair: Véronique François
Thursday 11:20 - 12:40 Public Transportation Room E503 - Chair: Javier Duran Micco
Thursday 11:20 - 12:40 Global Optimization Room E601 - Chair: Olivier Rigal
Thursday 11:20 - 12:40 Analytics 1 Room E602 - Chair: Rafael Van Belle
Thursday 14:10 - 15:10 Air, Rail and Multimodal Transportation Room E501 - Chair: Paola Pellegrini
Thursday 14:10 - 15:10 Game Theory Room E502 - Chair: Lotte Verdnock
Thursday 14:10 - 15:10 Transportation of People Room E503 - Chair: Yves Molenbruch
Thursday 14:10 - 15:10 Multi-level Optimization Room E601 - Chair: Concepcion Dominguez
Thursday 14:10 - 15:10 Analytics 2 Room E602 - Chair: Vedavyas Etikala
Thursday 15:50 - 16:50 Sport Timetabling Room E501 - Chair: Dries Goossens
Thursday 15:50 - 16:50 Project Management and Scheduling Room E502 - Chair: Fan Yang
Thursday 15:50 - 16:50 Rich Routing and Graphs Room E503 - Chair: Alexandre Bontems
Thursday 15:50 - 16:50 Logistics Room E601 - Chair: Silia Mertens
Thursday 15:50 - 16:50 Analytics 3 Room E602 - Chair: Jari Peeperkorn
Friday 11:20 - 12:40 Automatic Configuration and Metaheuristics Analysis Room E501 - Chair: Kenneth Sorensen
Friday 11:20 - 12:40 Real Life and Integrated Problems Room E502 - Chair: Jeroen Belien
Friday 11:20 - 12:40 Warehouse Management Room E601 - Chair: Harol Mauricio Gamez
Friday 11:20 - 12:40 Analytics 4 Room E602 - Chair: Noureddine Kouaissah
Friday 14:10 - 15:10 Orbel Award Room E501 - Chair: Jeroen Belien
Friday 14:10 - 15:10 Optimization Room E502 - Chair: Julien Dewez
Friday 14:10 - 15:10 Trucking Optimization Room E601 - Chair: Hatiz Çalik
Friday 14:10 - 15:10 Analytics 5 Room E602 - Chair: Diego Olaya
Friday 15:50 - 16:50 Learning and Optimization Room E501 - Chair: Jorik Jooken
- A Deep Reinforcement Learning approach to the Stochastic Inventory Problem
Henri Dehaybe (UCLouvain) Co-authors: Philippe Chevalier and Daniele Catanzaro
- Learning to optimize: a case study on a crane scheduling problem
Jorik Jooken (KU Leuven) Co-authors: Pieter Leyman, Patrick De Causmaecker, Tony Wauters Abstract: In many combinatorial optimization problems, the decision variables are all discrete. Hence, the set of all feasible solutions for a given problem instance can be represented by the leaf nodes of a tree in which a value is assigned to a decision variable at every level. In many cases, this tree is too large to be completely explored to find the best solution. However, one can still obtain very good solutions by not completely exploring the tree, but rather focusing on exploring the most promising parts of the tree. An example of an algorithm that uses this idea is Monte Carlo tree search. This algorithm has already been very successful in the context of game theory, where the most notable example is AlphaGo: a computer program that was able to defeat the world champion at the board game Go.
In this presentation we demonstrate that the core principles of Monte Carlo tree search for game theory can also successfully be applied in the context of a combinatorial optimization problem. The problem on which the developed methodology was tested, is a crane scheduling problem with a non-crossing constraint. The specific variant that we have concentrated on in this presentation is the variant 1D||C_Max according to the classification scheme proposed by Boysen et al. In this problem, the goal is to assign cranes to containers, stored along a straight line, in order to minimize the makespan (the latest completion time of a crane).
The tree search algorithm that is discussed in this presentation can be seen as a sampling method that samples the set of all possible solutions in a biased way (it is more likely to explore promising parts of the tree). To determine how exactly a sample will be generated, the algorithm uses two different policies: a rollout policy and a solution completion policy. The rollout policy is a stochastic policy that has to be learned by the algorithm. The solution completion policy on the other hand is a deterministic heuristic. To learn the rollout policy, the algorithm makes use of policy learning and this work hereby also contributes to recent attempts that focus on integrating machine learning into a combinatorial optimization context.
Friday 15:50 - 16:50 Lot-sizing and Inventory Room E502 - Chair: Philippe Chevalier
Friday 15:50 - 16:50 Rich Routing Room E601 - Chair: Cristian Aguayo
|
|
|
ORBEL - Conference chairs: Diego Cattaruzza and Maxime Ogier
|
|
|
|