1 présentation

• 
  10h30 - 10h55

  Data-driven optimization: structure detection, empirical optimality conditions, and instance-guided approaches.

  • Samir Elhedhli, prés., University of Waterloo

  We discuss three approaches for data-driven optimization where information, derived from data, is exploited in the design of efficient solution methodologies for hard optimization problems: detecting matrix structure, identifying empirical optimality conditions, and exploiting instance-level information. In the first, we use community detection to identify bordered block diagonal structure in constraint matrices of integer programs that allows the use of column generation/Lagrangian relaxation/branch-and-price methodologies. In the second, we identify an important characteristic of optimal solutions of hub location problems, spatial separability, and use it to solve very large problems in brain connectivity networks. Finally, we use data at the instance level to reduce the complexity of the underlying problem and design an efficient solution methodology for mixed-case palletization problems