Learning and Intelligent OptimizatioN conference

The high-level control of dynamic systems, such as aircraft, airports, air tra c, or spacecraft, consists in deciding at each control step on which action(s) to be performed as a function of current observations and objectives. Successive decisions must entail that the dynamics of the controlled system satis es user objectives as best as possible. To do so, a usual approach, inspired from the Model Predictive Approach in Automatic Control consists at each control step in (i) collecting current observations and objectives (ii) solving a deterministic planning problem over a given horizon ahead, (iii) extracting the rst action from the best plan produced, (iv) applying it, and (v) considering the next step. From the optimization point of view, this implies to be able to solve quickly many successive similar planning problems over a sliding horizon, maybe not in an optimal way. I will try to present and illustrate this approach and to explain the potential impact of learning techniques. Short bio: Graduated from école Polytechnique (Paris) in 1971 and from SUPAéRO (French national engineering school in aeronautics and space, Computer science specialization, Toulouse) in 1985, Gérard Verfaillie is now Research supervisor at ONERA (The French Aerospace Lab). His research activity is related to models, methods, and tools for combinatorial optimization and constrained optimization, especially for planning and decision-making. Autonomous Search Frédéric Saubion Université d'Angers, France Abstract: Decades of innovations in combinatorial problem solving have produced better and more complex algorithms. These new methods are better since they can solve larger problems and address new application domains. They are also more complex, which means that they are hard to reproduce and often harder to ne tune to the peculiarities of a given problem. This last point has created a paradox where e cient tools became out of reach for practitioners. Autonomous search represents a new research eld de ned to precisely address the above challenge. Its major strength and originality consist in the fact that problem solvers can now perform self-improvement operations based on analysis of the performances of the solving process including short-term reactive recon guration and long-term improvement through self-analysis of the performance, o ine tuning and online control, and adaptive control and supervised control. Autonomous search "crosses the chasm" and provides engineers and practitioners with systems that are able to autonomously self-tune their performance while e ectively solving problems. In this talk, we review existing works and we attempt to classify the di erent paradigms that have been proposed during past years to build more autonomous solvers. We also draw some perspectives and futures directions. Short bio: Frédéric Saubion coheads the Metaheuristics, Optimization and Applications team at the Université d'Angers (France); his research topics include hybrid and adaptive evolutionary algorithms and applications of metaheuristics to various domains such as information retrieval, nonmonotonic reasoning and biology. www.info.univ-angers.fr/pub/saubion Decades of innovations in combinatorial problem solving have produced better and more complex algorithms. These new methods are better since they can solve larger problems and address new application domains. They are also more complex, which means that they are hard to reproduce and often harder to ne tune to the peculiarities of a given problem. This last point has created a paradox where e cient tools became out of reach for practitioners. Autonomous search represents a new research eld de ned to precisely address the above challenge. Its major strength and originality consist in the fact that problem solvers can now perform self-improvement operations based on analysis of the performances of the solving process including short-term reactive recon guration and long-term improvement through self-analysis of the performance, o ine tuning and online control, and adaptive control and supervised control. Autonomous search "crosses the chasm" and provides engineers and practitioners with systems that are able to autonomously self-tune their performance while e ectively solving problems. In this talk, we review existing works and we attempt to classify the di erent paradigms that have been proposed during past years to build more autonomous solvers. We also draw some perspectives and futures directions. Short bio: Frédéric Saubion coheads the Metaheuristics, Optimization and Applications team at the Université d'Angers (France); his research topics include hybrid and adaptive evolutionary algorithms and applications of metaheuristics to various domains such as information retrieval, nonmonotonic reasoning and biology. www.info.univ-angers.fr/pub/saubion