With his teammates from the Technishe Universität München, Leo Baty won the competition organized jointly by NeurIPS (one of the 3 most selective machine learning conferences) and EURO (the European Society for Operations Research) on a dynamic vehicle routing problem. 54 teams of researchers from all over the world participated in this competition. Léo Baty is an alumni from Ecole des Ponts who is doing his thesis at Cermics, as part of the Air France chair, under the direction of Axel Parmentier.
The challenge was to produce an algorithm for the following routing problem. An express delivery company wants to minimize the distance driven by its vehicules. The requets arrive in real time. Every hour, the company decides which request to dispatch and builds the routes for the delivery vehicles. Requests must be delivered on time. A dilemma arises: should we wait for more request in the same area, or should we dispatch the request immediately to avoid having to make a last-minute delivery with an almost empty vehicle? This kind of dynamic and combinatorial problem is currently poorly solved. Building efficient algorithms can reduce the ecological and economical impacts of logistics systems. Leo Baty proposed an innovative algorithm based on techniques developed by the Cermics optimization team to combine machine learning and combinatorial optimization tools (see the associated preprint for more information: G. Dalle, L. Baty, L. Bouvier, & A. Parmentier, Learning with combinatorial optimization layers: A probabilistic approach, arXiv preprint 2207.13513).