Karell Bertet, Laboratoire L3i, Université de La Rochelle
Structure de Treillis : panorama des aspects structurels et algorithmiques.
Le premier ouvrage de référence de la théorie des treillis est la première édition du livre de Birkhoff en 1940. Cependant, la notion de treillis a été introduite dès la fin du 19ème siècle comme une structure algébrique munie de deux opérateurs appelés borne inférieure et borne supérieure. Depuis les années 2000, l’émergence de l’analyse formelle des concepts (FCA) dans divers domaines de l’informatique, que ce soit en analyse de données et classification, en représentation des connaissances ou en recherche d’information, a mis en avant les structures de treillis des concepts et de bases de règles d’implication. Plus récemment, les structures de patrons permettent d’étendre FCA à des données non binaires.
Une manipulation efficace de ces structures passe par une bonne connaissance du formalisme, des propriétés structurelles et des principaux résultats de la théorie des treillis et de l’AFC. Nous présenterons un panorama des concepts de base de la théorie des treillis, de FCA et des structures de patrons, ainsi que les principaux algorithmes de génération des objets qui la composent.
Karell Bertet est Maître de conférences au laboratoire L3i de l’université de La Rochelle depuis 1999, titulaire d’un doctorat en algorithmique de l’université Paris 7, et d’une habilitation en informatique de l’université de La Rochelle. Ses activités scientifiques ont pour fil directeur la structure de treillis, avec des résultats à la fois structurels, algorithmiques et applicatifs. Elle a publié plusieurs articles et encadré plusieurs thèses sur ce thème de recherche, et participe en particulier à l’animation de la communauté internationale de l’analyse formelle des concepts.
Antoine Cornuéjols, Professeur, AgroParisTech Paris
Apprentissage et classification par méthodes collaboratives : comment choisir ses collaborateurs et qu’échanger avec eux ?
Des problèmes d’intelligence artificielle aussi divers que les jeux à plusieurs joueurs, l’apprentissage supervisé par des méthodes d’ensemble ou le clustering collaboratif peuvent être considérées avec les mêmes questions fondamentales :
(1) comment choisir ses collaborateurs c’est-à-dire les sources d’information ?
(2) quelles informations il est intéressant d’échanger ?
et (3) comment combiner les informations reçues ?
Dans cet exposé nous examinerons successivement les algorithmes de jeux, un nouveau algorithme d’apprentissage supervisé par transfert puis le clustering collaboratif pour essayer d’en tirer des leçons sur les méthodes collaboratives pour la classification et, à tout le moins, de poser les bonnes questions.
Antoine Cornuéjols est professeur à AgroParisTech. Il est co-auteur avec Laurent Miclet et Vincent Barra de l’ouvrage « Apprentissage Artificiel. Deep learning, concepts et algorithmes ». Il s’intéresse particulièrement à l’apprentissage par transfert et aux méthodes d’apprentissage collaboratives supervisées et non supervisée.
Bernard De Baets, Professor, Ghent University
Monotonicity, a deep property in data science
In many modelling problems, there exists a monotone relationship between one or more of the input variables and the output variable, although this may not always be fully the case in the observed input-output data due to data imperfections. Monotonicity is also a common property of evaluation and selection procedures. In contrast to a local property such as continuity, monotonicity is of a global nature and any violation of it is therefore simply unacceptable. We explore several problem settings where monotonicity matters, including fuzzy modelling, machine learning and decision making. Central to the above three settings is the cumulative approach, which matches nicely with the monotonicity requirement.
By far the most popular fuzzy modelling paradigm, despite its weak theoretical foundations, is the rule-based approach of Mamdani and Assilian. In numerous applied papers, authors innocently assume that given a fuzzy rule base that appears monotone at the linguistic level, this will be the case for the generated input-output mapping as well. Unfortunately, this assumption is false, and we will show how to counter it. Moreover, we will show that an implication-based interpretation, accompanied with a cumulative approach based on at-least and/or at-most quantifiers, might be a much more reasonable alternative.
Next, we deal with a particular type of classification problem, in which there exists a linear ordering on the label set (as in ordinal regression) as well as on the domain of each of the features. Moreover, there exists a monotone relationship between the features and the class labels. Such problems of monotone classification typically arise in a multi-criteria evaluation setting. When learning such a model from a data set, we are confronted with data impurity in the form of reversed preference. We present the Ordinal Stochastic Dominance Learner framework, which permits to build various instance-based algorithms able to process such data.
Finally, we explore a pairwise preference setting where each stakeholder expresses his/her preferences in the shape of a reciprocal relation that is monotone w.r.t. a linear order on the set of alternatives. The goal is to come up with an overall monotone reciprocal relation reflecting `best’ the opinions. We formulate the problem as an optimization problem, where the aggregated linear order is that for which the implied stochastic monotonicity conditions are closest to being satisfied by the distribution of the input monotone reciprocal relations. Interesting links with social choice will be pointed out.
Bernard De Baets is a senior full professor in applied mathematics at the Faculty of Bioscience Engineering (Shanghai rank 37 in Life and Agriculture Sciences) of Ghent University, the top-ranked Belgian university (Shanghai rank 61). He is leading the research unit KERMIT and acts as head of the Department of Data Analysis and Mathematical Modelling. Furthermore, he is an affiliated professor at the Anton de Kom Universiteit (Suriname), an Honorary Professor of Budapest Tech (Hungary), a Doctor Honoris Causa of the University of Turku (Finland) and a Profesor Invitado of the Universidad Central “Marta Abreu” de las Villas (Cuba).
As a trained mathematician, computer scientist and knowledge engineer, Bernard has developed a passion for multi- and interdisciplinary research. He is not only deeply involved in fundamental research in three interlaced research threads, namely knowledge-based, predictive and spatio-temporal modelling, but he also aims at innovative applications in the applied biological sciences. At present, over 30 researchers are involved in the activities of KERMIT. Over the past 20 years, 75 PhD students have graduated under his (co-)supervision.
Bernard is a prolific writer, with a bibliography comprising over 500 peer-reviewed journal papers, 60 book chapters and 300 contributions to conference proceedings, accumulating more than 20000 Google Scholar citations (h-index 70). Several of his works have been bestowed upon with a best paper award. Moreover, he is a muchinvited speaker, having delivered over 250 lectures world-wide. In 2011, he was elected Fellow of IFSA (International Fuzzy Systems Association) and in 2012, he was a nominee for the Ghent University Prometheus Award for Research. In 2019, he received the EUSFLAT Scientific Excellence Award.
Bernard actively serves the research community, in particular as co-editor-in-chief of Fuzzy Sets and Systems and as member of the editorial board of several other journals, including the Internat. J. of Approximate Reasoning, Engineering Applications of Artificial Intelligence, and the Iranian J. of Fuzzy Systems. He is a member of the Administrative Board of the Belgian OR Society.
Dino Ienco, Chargé de recherche IRSTEA, UMR Tetis, Montpellier
Apprentissage de représentations avec des « connaissances faibles » : application au clustering et à la classification
Dans le contexte de la classification semi-supervisée, nous avons à notre disposition de gros volumes de données non étiquetées, mais très peu de données étiquetées et de connaissances associées aux données. Ce scénario n’est pas propice aux techniques d’apprentissage profond.
Dans cette présentation je vais illustrer mes derniers travaux de recherche en apprentissage semi-supervisé en utilisant des techniques d’apprentissage profond, avec des applications au clustering ainsi qu’à la classification transductive.
Alexandre Termier, Professeur, IRISA/Inria Rennes
Discovering habits with periodic patterns
In various fields, traces of timestamped events are captured, precisely describing the operation of a system or a process. It can as well be a web server log, the operation log of a manufacture, or even a personal activity journal. It can be interesting to analyze such data to discover periodic patterns, i.e. sets or sequences of events that occur within a regular delay. Two problems arise: first, precisely defining such patterns is non-trivial, as many straightforward definitions break in practice due to noise. Second, the search space of periodic patterns is huge, requiring some effort to make the enumeration efficient, and to output only the most interesting patterns. This talk will present the general problem of mining periodic patterns, our work to represent periodic patterns as formal concepts, and our latest work to use MDL approaches to select few relevant periodic patterns.
Alexandre Termier is Professor of Computer Sciences at Rennes 1 University, where he leads the Lacodam joint Inria-IRISA research group. His main research interest is pattern mining, both for broadening its application and for discovering new algorithms.
