Integer Linear Programming for Pattern Set Mining; with an Application to Tiling

Abdelkader Ouali (1) , Albrecht Zimmermann (2, 3) , Samir Loudni (2) , Yahia Lebbah (1) , Bruno Crémilleux (2, 3) , Patrice Boizumault (2) , Lakhdar Loukil (1)

1 LITIO - Laboratoire d'Informatique et Technologies de l'Information d'Oran
2 Equipe CODAG - Laboratoire GREYC - UMR6072
3 GREYC - Groupe de Recherche en Informatique, Image et Instrumentation de Caen

Pattern set mining is an important part of a number of data mining tasks such as classification, clustering, database tiling, or pattern summarization. Efficiently mining pattern sets is a highly challenging task and most approaches use heuristic strategies. In this paper, we formulate the pattern set mining problem as an optimization task, ensuring that the produced solution is the best one from the entire search space. We propose a method based on integer linear programming (ILP) that is exhaustive, declarative and optimal. ILP solvers can exploit different constraint types to restrict the search space, and can use any pattern set measure (or combination thereof) as an objective function, allowing the user to focus on the optimal result. We illustrate and show the efficiency of our method by applying it to the tiling problem.

Domaines: Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]

Abdelkader Ouali, Albrecht Zimmermann, Samir Loudni, Yahia Lebbah, Bruno Crémilleux, et al.. Integer Linear Programming for Pattern Set Mining; with an Application to Tiling. Pacific-Asia Conference on Knowledge Discovery and Data Mining, May 2017, Jeju, South Korea. ⟨10.1007/978-3-319-57529-2_23⟩. ⟨hal-01597819⟩