The EURO Gold medal of the Association of European Operational Research Societies (EURO) is the most important European scientific prize for operations research. Martine is the first woman to have received this prize.


Place: Inria LilleNord Europe
Abstract: We have all experimented recommender systems on the web, whether it’s for choosing a movie, a book, a restaurant, or a holiday destination. Roughly speaking, a recommender system is a platform that seeks to predict the rating or preference that a user would give to an item.
More formally, consider a set 𝐼 of items and a set 𝐴 of Boolean attributes. A Boolean vector 𝒙𝑖 with 𝐴 components can be associated with every item 𝑖 in 𝐼 so that the 𝑗th component of 𝒙𝑖 equals 1 if and only if the 𝑗th attribute in 𝐴 is true for item 𝑖. For example, if 𝐼 is a set of restaurants and the first attribute is “vegetarian”, then the first component of vector 𝒙𝑖 associated with restaurant 𝑖 in 𝐼 equals 1 if and only if 𝑖 is a restaurant offering vegetarian food.
Consider now a set 𝑈 of users. A Boolean vector 𝒚𝑢 with A components can be associated with every user 𝑢 in 𝑈 so that the 𝑗th component of 𝒚𝑢 equals 1 if and only if user 𝑢 has interest for the 𝑗th attribute. In our example, the first component of vector 𝒚𝑢 associated with a user 𝑢 in 𝑈 equals 1 if and only if user 𝑢 has interest for vegetarian restaurants. While these vectors 𝒚𝑢 are not known, we show how resolving sets can predict them, which makes it to possible to recommend to user 𝑢 all items 𝑖 with a vector 𝒙𝑖 closest to 𝒚𝑢.
We say that a vector 𝒂 resolves two vectors b and c if the Hamming distance between 𝒂 and b is different from that between 𝒂 and c. A subset 𝐼′ of items is a resolving set if every two distinct vectors 𝒚𝑢 and 𝒚𝑢′ are resolved by at least one vector 𝒙𝑖 with 𝑖 in 𝐼′.
Given a resolving set 𝐼′={𝑖1,…, 𝑖𝑟} with 𝑟 items, we can associate a vector 𝒛𝑢 with 𝑟 integer components to every user 𝑢 so that the 𝑘th component of 𝒛𝑢 is the Hamming distance between 𝒙𝑖𝑘 and 𝒚𝑢. As a consequence, given two users 𝑢 and 𝑢′, 𝒛𝑢 not equal to 𝒛𝑢′ if and only if 𝒚𝑢 not equal to 𝒚𝑢′. In other words, it is possible to differentiate between two users with distinct preferences on the basis of their distances to the vectors 𝒙𝑖 with 𝑖 in 𝐼′.
We show in this talk that it is easy to determine resolving sets of small size. For example, for 20 attributes (which allows the classification of the items in more than one million categories), resolving sets with 12 items can be determined. In other words, it is sufficient to determine the 12 components of 𝒛𝑢 for every user 𝑢 to predict the preference vector 𝒚𝑢 of user 𝑢 among more than one million item types.
Place: Université Libre de Bruxelles
Abstract: Optimization models for Air Traffic Flow Management (ATFM) select, for each flight, suitable trajectories with the aim of reducing congestion of both airports and enroute sectors, and maximizing the Air Traffic Management system efficiency. Recent models try to include as accurate as possible information on airspace capacity as well as on Airspace Users route preferences and priorities, as suggested by recent SESAR (Single European Sky ATM Research) programs. In this direction, we analyze flight trajectories queried from Eurocontrol DDR2 data source. We learn homogeneous trajectories via clustering, and we apply data analytics (mainly based on tree classifiers, support vector machines and multiple regression) to explore the relation between grouped trajectories and potential choicedeterminants such as length, time, enroute charges, fuel consumption, aircraft type, airspace congestion etc. The associations are evaluated and the predictive value of determinants is validated and analyzed. For any given origindestination pair, this ultimately leads to determining a set of flight trajectories and information on related Airspace Users’ preferences that feed optimization models for ATFM.
]]>Place: Inria LilleNord Europe
Abstract: Demand response is a valuable resource in the smart grid. It has been proved that demand response can improve the reliability of the grid while decreasing load consumption at peak time. Aggregators play a major role in the smart grid: as intermediaries between the utility and the customers, they have to satisfy the requests of the utility while meeting the customers’
constraints. Demand response requests have to be distributed optimally by the aggregators to improve the reliability of the grid at peak time. The uncertainty of the availability of the resources has to be included in the optimization model in order to satisfy the request of the utility. We propose to include this uncertainty in the optimization model using the chanceconstrained method. In order to have an accurate optimal solution, aggregators predict the load consumption over a shortterm horizon, and commit sufficient demand response requests at minimum cost while ensuring sufficient backup capacity to mitigate the uncertainty. We observe that support vector regression offers a good shortterm forecast for the loads in the residential and commercial sector.
This is joint work with Miguel F. Anjos, Laurent Lenoir, and Dalal Asber.
]]>Place: Université Libre de Bruxelles
Abstract: Hubbing is commonly used in airlines, cargo delivery and telecommunications networks where traffic from many origins to many
destinations are consolidated at hubs and are routed together to benefit from economies of scale. Each application area has its own specific features and the associated hub location problems are of complex nature. In the first part of this talk, I will introduce
the basic hub location problems, summarize important models and results and mention the shortcomings of these in addressing real
life situations. In the second part, I will introduce new variants of the hub location problem that incorporate features such as
hierarchical and multimodal networks, service of quality constraints, generalized allocation strategies and demand uncertainty. I will conclude the talk with an ongoing work on a joint problem of hub location, network design and dimensioning.
Place: Université Libre de Bruxelles
Abstract: In this talk, we present diverse concepts: Variable Neighborhood Search, Variable Neighborhood Descent, parallel algorithm, shared memory, message passing interface and community detection. Besides, we explain how can we use all of them in only one algorithm. Some numerical results will be discussed at the end.
]]>Place: Inria LilleNord Europe
Abstract: This seminar will show mathematical models and algorithms for optimizing the management of crews and aircrafts of an airline company operating flights in Canary Islands. It is the result of a research work for a company, and it can be extended to be
applied to other transportation companies with similar characteristics. The seminar will concern with the daily planning and the crew rostering problems, although our research has also considered more complicated issues like management disruption.
The main part of the seminar will introduce, model and solve a new vehicle routing problem arising when planning the sequence of legs that each crew and aircraft of an airline company must perform. It was motivated by a realworld problem in Canary Island, where the airline operates flights between 11 airports. There are no flight during the night, there are about 180 flights (legs) during a day, and the flying time between two airports is around 30 minutes. The crews of the airline company live in the two major islands (Tenerife and Gran Canaria). The airport of Gran Canaria concentrates all the equipment to perform the maintenance of the aircrafts,
which must be performed immediately after two operating days. Thus, an aircraft starting the journey outside Gran Canaria must finish in Gran Canaria. Instead, each crew is expected to return to the island where it started that day. There is no limitation on the number of legs that an aircraft can fly in a day, but the number of legs assigned to a crew in a day is limited by law. There is also a limitation in the activity time of each crew in a day. The aim of the problem is to find a sequence of legs to each crew and each aircraft minimizing a cost function while satisfying the above constraints. The above problem can be seen as a routing problem with two depots, where one must find routes for two type of vehicles (crews and aircrafts). We give a mathematical formulation of the whole routing problem and a branchandcut algorithm to solve it. Note that it integrates the assignment, aircraft routing and crew schedulling. We will also discuss the crew and aircraft rostering. The performances of our implementation is evaluated and discussed on realworld instances.
This seminar will show results that have been published in two articles. One published in the journal OMEGA ( doi:10.1016/j.omega.2013.06.006 ) and another in the journal TRANSPORTATION SCIENCE ( doi:10.1287/trsc.2015.0655 )
]]>Place: ULB Bruxelles
Abstract: In this talk we introduce an stochastic optimization/gametheory model to describe an energy market working on a network (transmission), including renewal energy suppliers, distributed generators and demand nodes. We characterize Nash equilibria and pricing rules in this stochastic setting as well as some stability properties and market power behaviors.
]]>