Search

International audience; Exceptional Model Mining (EMM) is a local pattern mining framework that generalizes subgroup discovery. In EMM, we look for subsets of objects-subgroups-whose model deviates significantly from the same model fitted on the overall dataset. Multi-objective Optimization (MOO) is an area of Multiple Criteria Decision Making where two or more functions need to be optimized at the same time and the goal is to find the best compromise between the concurrent objectives. We introduce a new model class for EMM in a MOO setting called Exceptional Pareto Front Mining. We design fitting quality measures that take into account both the distance between models and the relevance of the subgroups. We propose a beam search for top-K EMM whose added-value is studied on both synthetic and real life datasets. Among others, we discuss a use case on hyperparameter optimization in machine learning for both regression and multi-label classification.

International audience; Subgroup discovery in labeled data is the task of discovering patterns in the description space of objects to find subsets of objects whose labels show an interesting distribution, for example the disproportionate representation of a label value. Discovering interesting subgroups in purely numerical data-attributes and target label-has received little attention so far. Existing methods make use of discretization methods that lead to a loss of information and suboptimal results. This is the case for the reference algorithm SD-Map*. We consider here the discovery of optimal subgroups according to an interestingness measure in purely numerical data. We leverage the concept of closed interval patterns and advanced enumeration and pruning techniques. The performances of our algorithm are studied empirically and its added-value w.r.t. SD-Map* is illustrated.

International audience; Designing, selling and/or exploiting connected vertical urban farms is now receiving a lot of attention. In such farms, plants grow in controlled environments according to recipes that specify the different growth stages and instructions concerning many parameters (e.g., temperature , humidity, CO2, light). During the whole process, automated systems collect measures of such parameters and, at the end, we can get some global indicator about the used recipe, e.g., its yield. Looking for innovative ideas to optimize recipes, we investigate the use of a new optimal subgroup discovery method from purely numerical data. It concerns here the computation of subsets of recipes whose labels (e.g., the yield) show an interesting distribution according to a quality measure. When considering optimization, e.g., maximizing the yield, our virtuous circle optimization framework iteratively improves recipes by sampling the discovered optimal subgroup description subspace. We provide our preliminary results about the added-value of this framework thanks to a plant growth simulator that enables inexpensive experiments.