Skip to Content

PEPS Préfute (CNRS 2015-2016) - fouille interactive fondée sur les PREFérences UTilisatEur - Laboratoires CERMN (Caen), GREYC (Caen), LI (Blois/Tours), LIRIS (Lyon) et LORIA (Nancy)

The Decade project is partly a follow-up of the Prefute project.

 

Colloque Mastodons des 9 et 10 février 2017

La présentation Prefute.

 

Mini-symposium on instant data mining, interactive data mining, preference-based pattern mining (October 26-28, 2016). Joint event with the Mastodons HyQual project.

Venue: room - A picture of the audience
IRISA INRIA Rennes
12 F, allée Jean Perrin 35 042 Rennes
How to reach IRISA INRIA Rennes ?

This event is partly supported by CNRS, INRIA and French Research National Agency (funded project Hybride).

 

2016 October Wednesday 26th

 

  • 14h30: welcome
     
  • 14h45 "Some Recent Advances in Exceptional Model Mining - Unusual Preferences and more", Wouter Duivesteijn (Ghent University, Belgium & TU Eindhoven, The Netherlands).
    Abstract: My favorite form of pattern mining is Exceptional Model Mining (EMM): a supervised form of local pattern mining, where multiple attributes are singled out as the targets, and subgroups are deemed interesting if the targets interact in an unusual way.  In this talk, I will outline some new developments in the world of EMM. On the one hand, we will discuss Exceptional Preferences Mining, an instance of EMM where the relevant type of interaction is unusual preference behavior over a fixed set of labels. On the other hand, we will explore a model class which is compatible with the framework of subjective interestingness. Rather than objectively defining a single quality measure that will determine which subgroups are objectively interesting once and for all, the MEP (for: Most Extreme Projection) model class for EMM allows the user to express prior expectations about the data, and then seeks subgroups that are subjectively unusual with respect to the knowledge at hand. In exploring these two new aspects of EMM, we will encounter observations about the peculiar appreciation of certain types of sushi in Northern regions of Japan, and on parts of Europe that, having learned some things about them, cease to be subjectively interesting.
     
  • Discussion and break
     
  • 16h15 "Finding good sets of redescriptions using MDL, MaxEnt and interactive visualization", Esther Galbrun (INRIA-NGE, Nancy).
    Abstract: How to go from finding sets of good patterns to finding good sets of patterns? I'll focus on this issue in the context of redescription mining and present methods based on the minimum description length principle and maximum entropy principle that have been tailored for that class of patterns. I will also touch on techniques for interactively visualizing multiple redescriptions and how visualization and modeling could support each other (probably with more questions than answers).
     
  • "Which pattern to test next: witnesses and other ideas", Albrecht Zimmermann (GREYC, Caen).
    Abstract: When exploring the search space in a pattern mining operation, the quintessential goal is to test as few patterns as possible against the data. One idea that has been proposed (but not pursued by its creaters) is that of witnesses, patterns whose satisfaction of a constraint (or lack thereof) allow us to draw conclusions about a certain set of untested patterns. In this short talk, I'll sketch the witness idea and try to relate it to others, such as sampling the pattern space, propagation techniques in constraint programming, etc."
  • "Abstract and Local mining in Attributed Networks", Henry Soldano (LIPN, Paris).
    Abstract: Attributed networks, as encountered in social networks or life science, may be considered both as  sets of nodes described by labels belonging to some pattern language, and as a graphs whose edges relate these nodes. Pattern mining in attributed networks considers patterns constraining both the node subset and the connectivity, thus resulting in dense subnetworks whose nodes labels satisfy some  constraint. We propose to apply closed pattern mining ideas to this problem by considering the graph information as an external knowledge source that allows reducing the support sets of patterns to core subgraphs or their maximal connected parts  also called structural communities.

 

2016 October Thursday 27th

 

  • 9H "From Case Studies to High-Throughput CTFs---Addressing the Evaluation Bottleneck in KDD Research", Mario Boley (Max Planck Institute for Informatics, Saarbrücken, Germany).
    Abstract: You can't manage what you can't measure" is a slogan predominantly cited in the business sector. However, the ability to effectively evaluate performance is just as crucial for lasting success in research and engineering. This is witnessed by the predictive modeling community, where the availability of high-throughput common task frameworks (CTFs) lead to an impressive volume of published research with no apparent tendency to decline anytime soon. In contrast, the part of the KDD community that deals with the research and development of user-centric data analysis tools seems to be caught in a constant struggle to appropriately convince a wider audience (and sometimes even itself) of the significance of its contributions. In this talk, I review different forms of evaluation that, if adopted more systematically and consciously, could potentially alleviate this struggle. All forms have in common that they establish an external evaluation context (task). Initially, in case studies, this context can be purely anecdotal and exemplary and still convincingly motivate the development of new techniques or the necessity to alter existing approaches. Subsequently, as claims about the user/system interaction become more central to the contribution to be evaluated, studies involving real users are indicative. Through formalizing an increasing number of aspects of the evaluation context, a higher scalability of studies can be reached. Ultimately, of course, we would like to fully close the gap to high-throughput CTFs and replace measurement of real user behavior by predictions of abstract interaction models. I end with some ideas of how this final step could be achieved.
     
  • Discussion and break
     
  • 10h30: "Interactive discovery of user interests using pattern sampling", Arnaud Giacometti/Arnaud Soulet (LI, Tours).
    Abstract: Most pattern mining methods require an interestingness measure as input. In practice, the user is often unable to explicitly express his/her interest. But, as he/she will able to judge whether a given pattern is relevant or not, recent proposals focus on interactive discovery processes. In this paper, we propose a new interactive pattern mining method assuming that only a part of the dataset is really interesting for the user. By integrating user feedback about patterns, our method aims at sampling patterns with a probability proportional to their frequency in the preferred transactions. We demonstrate that our method accurately identifies this set of preferred transactions if feedback are consistent with user's interest. Experiments also show the good performances of our approach in terms of precision and recall.
     
  • "Any-time Diverse Subgroup Discovery with Monte Carlo Tree Search", Guillaume Bosc (LIRIS, Lyon).
    Abstract: Discovering descriptions that highly distinguish a class label from another is still a challenging task. Such patterns enable the building of intelligible classifiers and suggest hypothesis that may explain the presence of a label. Subgroup Discovery (SD), a framework that formally defines this pattern mining task, still faces two major issues: (i) to define appropriate quality measures characterizing the singularity of a pattern; (ii) to choose an accurate heuristic search space exploration when a complete enumeration is unfeasible. To date, the most efficient SD algorithms are based on a beam search. The resulting pattern collection lacks however of diversity due to its greedy nature. We propose to use a recent exploration technique, Monte Carlo Tree Search (MCTS). To the best of our knowledge, this is the first attempt to apply MCTS for pattern mining. The exploitation/exploration trade-off and the power of random search leads to any-time mining (a solution is available any-time and improves) that generally outperforms beam search. Our empirical study on various benchmark and real-world datasets  shows the strength of our approach with several  quality measures.
     
  • "Unsupervised Exceptional Attributed Sub-graph Mining in Urban Data", Ahmed Anes Bendimerad  (LIRIS, Lyon).
    Abstract: Geo-located social media provide a wealth of information that describes urban areas based on user descriptions and comments. Such data makes possible to identify meaningful city neighborhoods on the basis of the footprints left by a large and diverse population that uses this type of media. In this paper, we present some methods to exhibit the predominant activities and their associated urban areas to automatically describe a whole city. Based on a suitable attributed graph model, our approach identifies neighborhoods with homogeneous and exceptional characteristics. We introduce the novel problem of exceptional subgraph mining in attributed graphs and propose a complete algorithm that takes benefits from new upper bounds and pruning properties. We also propose an approach to sample the space of exceptional subgraphs within a given time-budget. Experiments performed on 10 real datasets are reported and demonstrate the relevancy and the limits of both approaches.
     

Lunch

  • 14h "Pattern compositions", Jilles Vreeken (Saarland University, Saarbrücken, Germany).
    Abstract: The goal of exploratory data analysis -- or, data mining -- is making sense of data. We develop theory and algorithms that help you understand your data better, with the lofty goal that this helps formulating (better) hypotheses. More in particular, our methods give detailed insight in how data is structured: characterising distributions in easily understandable terms, showing the most informative patterns, associations, correlations, etc. My talk will consist of two parts. I will start by explaining what is a pattern composition. Simply put, databases often consist of parts, each best characterised by a different set of patterns. Young parents, for example, exhibit different buying behaviour than elderly couples. Both, however, buy bread and milk. A pattern composition jointly characterises the similarities and differences between such components of a database, without redundancy or noise, by including only patterns that are descriptive for the data, and assigning those patterns only to the relevant components of the data. In the second part of my talk I will argue the generality of the framework, how can be used for a variety of data types, including sequences and graphs, how it connects to topic modelling and subgroup discovery, and, I will make the connection to causal discovery.
     
  • Discussion and break
     
  • 15h30  "Towards data science automation: some ideas", Alexandre Termier (IRISA, Rennes).
    Abstract: Data Science is the activity that consists in making sense from (large volumes of) data. It requires a lot of human expertise, and is usually performed by few highly skilled professionals, the “Data Scientists”. Their limited number means that few organizations can benefit from their expertise. Due to the huge need for better data analysis, large efforts are beginning to automate as much of the data science process as possible. In this talk, we will present some ideas that we are working on to automate parts of this process.
     
  • "Pattern-based progressive analytics on user trace logs" Julien Blanchard (LINA, Nantes).
    Abstract: Looking for user behaviors in interaction trace logs is a difficult problem. Firstly, because this is an exploratory process, in which the analyst's subjectivity plays a major role. And secondly, because the traces are generally too detailed to allow the analyst to read behaviors inside them directly. In this talk, we present our first proposals towards developing tools allowing an analyst to discover user behaviors in traces by interacting with both pattern mining algorithms and information visualizations in a progressive analytics way.
     
  • "A Framework for Actionable Clustering using Constraint Programming", Thi-Bich-Hanh Dao (LIFO, Orléans).
    Abstract: Consider if you wish to cluster your ego network in Facebook so as to find several useful groups each of which you can invite to a different dinner party. You may require that each cluster must contain equal number of males and females, that the width of a cluster in terms of age is at most 10 and that each person in a cluster should have at least r other people with the same hobby. These are examples of cardinality, geometric and density requirements/constraints respectfully that can make the clustering useful for a given purpose. However existing formulations of constrained clustering were not designed to handle these constraints since they typically deal with low-level, instance-level constraints. We formulate a constraint programming (CP) languages formulation of clustering with these cluster-level styles of constraints which we call actionable clustering. Experimental results show the potential uses of this work to make clustering more actionable. We also show that these constraints can be used to improve the accuracy of semi-supervised clustering.

 

2016 October Friday 28th

 

  • 9H "IoT Big Data Stream Mining", Albert Bifet (Télécom ParisTech, Paris).
    Abstract: Big Data and the Internet of Things (IoT) have the potential to fundamentally shift the way we interact with our surroundings. The challenge of deriving insights from the Internet of Things (IoT) has been recognized as one of the most exciting and key opportunities for both academia and industry. Advanced analysis of big data streams from sensors and devices is bound to become a key area of data mining research as the number of applications requiring such processing increases. Dealing with the evolution over time of such data streams, i.e., with concepts that drift or change completely, is one of the core issues in stream mining. In this talk, I will present an overview of data stream mining, and I will introduce some popular open source tools for data stream mining.
     
  • Discussion and break
     
  • 10h30 "Framester: A Wide Coverage Linguistic Linked Data Hub", Mehwish Alam (LIPN, Paris).
    Abstract: Semantic web applications leveraging NLP can benefit from easy access to expressive lexical resources such as FrameNet. However, the usefulness of FrameNet is affected by its limited coverage and non-standard semantics. The access to existing linguistic resources is also limited because of poor connectivity among them. We present some strategies based on Linguistic Linked Data to broaden FrameNet coverage and formal linkage of lexical and factual resources. We created a novel resource, Framester, which acts as a hub between FrameNet, WordNet, VerbNet, BabelNet, DBpedia, Yago, DOLCE-Zero, as well as other resources. Framester is not only a strongly connected knowledge graph, but also applies a rigorous formal treatment for Fillmore's frame semantics, enabling full-fledged OWL querying and reasoning on a large frame-based knowledge graph. We also describe Word Frame Disambiguation, an application that reuses Framester data as a base in order to perform frame detection from text, with results comparable in precision to the state of the art, but with a much higher coverage.
     
  • "Contribution to the classification of web of data based on Formal Concept Analysis and Pattern Structures", Justine Reynaud (LORIA, Nancy).
    Abstract: The emergence of Web of Data enables the production of very large ontologies and RDF datasources such as Yago or DBpedia. These ontologies can be connected together, constituting the Linked Open Data (LOD) cloud. Basic units of these ontologies are RDF triples expressed as (subject, predicate, object). One issue of main interest is knowledge discovery within LOD, which can help information retrieval and knowledge engineering. FCA and its extensions, such as Pattern Structures (PS), were already used to classify LOD elements. Here, we will present a specific pattern structure used to mine definitions in DBPedia from a set of triples with the use of background knowledge.
     
  • "Mining over uncertain databases", Ahmed Samet (IRISA, Rennes).
    Abstract: In recent years, the mining of frequent itemsets from uncertain databases became a hot topic within data mining community. In contrast with binary databases where extraction is a deterministic problem, the uncertain case relies on expectation. Recently, a new type of databases that models both uncertain and imprecise data has emerged and called evidential database. In this short talk, we present an applicative study case of evidential databases in chemistry. Then, we shed light on a predictive model for amphiphile molecule properties.
     
  • 11H30: closing discussion
     

Lunch

 

 

Journée du 25 août 2016 (Paris)

  • matin : travail sur tutoriel "Preference-based Pattern Mining"
  • 14h00 : Prelinda

 

Lieu

Salle de réunion Gilles Kahn 1 - Bâtiment C - Etage 0
Centre de recherche INRIA Paris
2 rue Simone Iff 75012 Paris
Commet venir à INRIA Paris ?
 

 

Session commune avec le PEPS Hydrata (HYpergraphes et Datamining : AlgoriThmes et Analyses probabilistes) - 7 juin 2016 (Paris)

En savoir plus sur le PEPS Hydrata

  • 09h30-10h30 : base minimale d'implications et traverses minimales (Alexandre Bazin, ISIMA)
  • 10h30-11h00 : pause
  • 11h00-12h00 : vers une mesure de similarité pour les séquences complexes (Elias Egho, Orange Lab). Résumé : le calcul de similarité entre les séquences est d’une extrême importance dans de nombreuses approches d’explorations de données. Il existe une multitude de mesures de similarités de séquences dans la littérature. Or, la plupart de ces mesures sont conçues pour des séquences simples, dites séquences d’items. Dans ce travail, nous étudions d’un point de vue purement combinatoire le problème de similarité entre des séquences complexes (i.e., des séquences d’ensembles ou itemsets). Nous présentons de nouveaux résultats afin de compter efficacement toutes les sous-séquences communes à deux séquences. Ces résultats théoriques sont la base d’une mesure de similarité calculée efficacement grâce à une approche de programmation dynamique.
  • Déjeuner au CROUS Buffon.

 

Lieu

Bâtiment Sophie Germain - Salle 0011 - Université Paris 7
(bâtiment du labo IRIF, ex-LIAFA)
8 place Aurélie Nemours
75013 Paris
Metro 14 - RER C - Bibliothèque François-Mitterrand - Accès

 

Journée du 6 juin 2016 (Paris)

  • 9h45 : introduction
  • 10h : recherche d’ensembles de zones géographiques connexes et exceptionnelles : vers une approche instant data mining ? (Anes Bendimerad, LIRIS)
  • Quelques perspectives sur la génération de données (Albrecht Zimmermann, GREYC)
  • Classification via la médiane (LORIA)
  • Déjeuner
  • 14h00 : En 2 slides :  scénario pour la découverte de motifs chimiques caractéristiques d’une activité biologique (Bruno Crémilleux, GREYC)
  • Démo :  Interactive Discovery of Hypotheses on the Structure-Odor Relationship in Neuroscience (Guillaume Bosc, LIRIS)
  • discussion : préparartion du workshop, activités intra et inter-partenaires lieés à Préfute, liens avec Approppre et HyQual
  • 15h00 : préparation de Prelinda
  • Diner : restaurant l'Ecumoir - 64 Rue Saint-Sabin, 75011 Paris.

 

Lieu

Salle de réunion Gilles Kahn 2 - Bâtiment C - Etage 0
Centre de recherche INRIA Paris
2 rue Simone Iff 75012 Paris
Commet venir à INRIA Paris ?
 

 

Mini-symposium on instant data mining and learning preferences for interactive data mining. Joint event with the PEPS Approppre - November 19-20th (Nancy)

 

2015 November Thursday 19th:

 

  • 9h15: welcome
     
  • 9h30: "Mine, Interact, Learn, Repeat" Matthijs van Leeuwen (Leiden University, The Netherlands).
    Abstract: Both data and data analysis techniques are ubiquitous, but most techniques focus on solving a single task and a different algorithm is required for each specific task. Consequently, exploring data and discovering interesting, novel insights can be challenging, in particular because data exploration algorithms are generally unable to learn what you find interesting. In this talk, I will show how techniques from exploratory data mining and machine learning can be combined to establish exploratory mining algorithms that learn what the user finds interesting. As suggested by the title of this talk, this can be achieved by first Mining patterns, then letting the user Interact with these intermediate results, Learning about subjective interestingness from the user feedback, and Repeating these steps.
     
  • 10h15: discussion and break
     
  • 11h00: "Anytime Algorithm for Frequent Pattern Outlier Detection" (Arnaud Giacometti and Arnaud Soulet, LI).
    Abstract: Outlier detection consists in detecting anomalous observations. Recently, outlier detection methods have proposed to mine all frequent patterns in order to compute the outlier factor of each transaction. We propose an anytime algorithm for frequent pattern outlier detection based on pattern sampling. This approach not only provides a good approximate solution at any time, but also estimates the error thanks to Bennett’s inequality.
     
  • 11h30: "Interactive Knowledge Discovery over Web of Data" (Mehwish Alam, LORIA).
    Abstract: Recently, the "Web of Documents" has become the "Web of Data", i.e., the documents are annotated in the form of RDF (Resource Description Framework) making this human processable data directly processable by machines. This data can further be explored by the user using SPARQL queries. As web clustering engines provide classification of the results obtained by querying web of documents, a framework for providing classification over SPARQL query answers is also needed to make sense of what is contained in the data. Exploratory Data Mining focuses on providing an insight into the data. It also allows filtering of non-interesting parts of data by directly involving the domain expert in the process. This thesis contributes in aiding the user in exploring Linked Data with the help of exploratory data mining. We study three research directions, i.e., 1) Creating views over RDF graphs and allow user interaction over these views,  2) assessing the quality and completing RDF data and finally 3) simultaneous navigation/exploration over heterogeneous and multiple resources present on Linked Data. Firstly, we introduce a "solution modifier" i.e., "View By" to create views over RDF graphs by classifying SPARQL query answers with the help of Formal Concept Analysis. In order to navigate the obtained concept lattice and extract knowledge units, we develop a new tool called RV-Xplorer (Rdf View eXplorer) which implements several navigational modes. However, this navigation/exploration reveal several incompletions in the data sets. In order to complete the data, we use association rule mining for completing RDF data.  Furthermore, for providing navigation and exploration directly over RDF graphs along with background knowledge, RDF triples are clustered w.r.t. background knowledge and these clusters can then be navigated and interactively explored. Finally, it can be concluded that instead of providing direct exploration we use FCA as an aid for clustering RDF data and allow user to explore these clusters of data and enable the user to reduce his exploration space by interaction.
     
  • Lunch
     
  • 14h: "Subjective interestingness and interactive data mining: two sides of the same coin?" Tijl de Bie (Ghent University, Belgium).
    Abstract: A key aspect of exploratory data mining is the quantification of the "interestingness" of a data mining pattern. During the first part of this talk I will survey our ongoing research on building a framework within which it is possible to quantify the interestingness of patterns found in data in a subjective manner. This framework explicitly considers exploratory data mining as an iterative process. During the second part of this talk, I will then discuss how also interactions by the data analyst can naturally be accommodated within this framework. I will conclude with some suggestions for further research on the interface between subjective interestingness and interactivity in data mining.
     
  • 14h45: dicussion and break
     
  • 15h30: "Exceptional Local Model Mining and its Ability to Elicit Hypothesis in Olfaction" Guillaume Bosc (LIRIS).
    Abstract: From a molecule to the brain perception, the olfaction is a complex phenomenon that remains to be fully understood in neuroscience. One challenge is to establish comprehensive rules between the physicochemical properties of the molecules (weight, atom counts,...) and specific and small subsets of olfactory qualities (fruity, woody,...). Subgroup discovery makes it possible to find such descriptive rules. However, existing methods are able to characterize either a single or all labels (exceptional model mining). Moreover, classical quality measures (weighted relative accuracy, weighted Kullback-Leibler divergence or the F-Score) are too strict w.r.t. the recall or the precision, to characterize both over- and under-represented label subsets. To tackle these problems, we propose an original subgroup discovery approach where search space exploration, quantitative attributes discretization and quality measure definition need to be jointly revisited. Our experiments on an olfactory database provide intelligible, non-redundant and statistically significant subgroups that strongly discriminate small fragrant sets.
     
  • 16h: "A dimensionality reduction method for qualitative preference aggregation" Quentin Brabant (LORIA).
    Abstract: We consider a qualitative framework for supervised preference learning problem in which alternatives are described according to n criteria and evaluated in an ordered set, not necessarily linearly ordered, by a global utility function that aggregates the evaluations on each criteria. Such a learning problem then reduces to learning the aggregation model. In this ordinal setting, a suitable aggregation model is the Sugeno integral. Learning such a model with ordinal data can thus be turned into an optimization problem with 2^n parametrers,  that easily becomes an intractable problem as the number of criteria increases. In this talk we propose a method for reducing the number of criteria while keeping reasonable accuracy when the evalutation space is a bound distributive lattice, and present some preliminary experimental results.
     
  • 16h30: "Preference-based optimal patterns" Bruno Crémilleux et al. (GREYC).
    Abstract: User preferences can be very helpful for specifying pattern mining problems. At the same time, there are also different "families" that specify "relations" among patterns, such as the closed patterns, top-k patterns, pattern sets etc. This talk addresses the question of how user preferences and different pattern families relate to each other. Starting from the notion of preference, we introduce Optimal Patterns (OPs) which are the best patterns according to a given user preference. We show how model and mine OPs thanks to the Constraint Programming paradigm, and put OPs into the context of pattern families.
  • Discussion

 

2015 November Friday 20th:

  • 10h00: "When cyberathletes conceal their game: clustering confusion matrices to identify avatar aliases" Victor Codocedo (LORIA).
  • 11h00-13h00:  working groups and discussions.

 

Venue:

LORIA lab, room C005 (19/11) - B013 (20/11)
How to reach the LORIA?

 

Journée du 29 septembre 2015 (Caen)

  • 9h00 : accueil et introduction
  • 9h15 - 9h45 : "Preference-based skypatterns" (S. Loudni, Caen)
  • 9h45 - 10h15 : "Constraint Acquisition" (N. Lazaar, Montpellier)
  • 10h15 - 10h45 : "Découverte instantanée de motifs" (A. Soulet, Tours)
  • Pause
  • 11h00 - 12h15 : Présentation des PEPS Apostrop et Préfute et discussion
  • Déjeuner
  • 14h00 - 16h00 : Discussion et perspectives

 

Lieu

GREYC - Salle S3 351
UFR Sciences
Université Caen Normandie
Plans d'accès
 

Session commune avec le PEPS Approppre (Approche ordinale pour l'apprentissage et la prédiction de préférences) - 7 juillet 2015 (Paris)

  • 10h15 : accueil
  • 10h30 - 11h00 : "Some Elements on Preferences and Formal Concept Analysis" (Amedeo)
  • 11h00 - 11h30 : "A characterization of the 2-additive Choquet integral by using binary alternatives " (Brice)
  • 11h30 - 12h00 : "Application des intégrales de Sugeno k-maxitives à la modélisation de préférences multicritères" (Quentin )
  • 12h00 - 13h00 : Discussion et perspectives
  • Début de réunion Approppre + Prefute
  • 14h : introduction
  • 14h30 - 15h00 : "Mathematical Morphology on Complete Lattices for Imperfect Information Processing"(Isabelle)
  • 15h00 - 15h30 : "Mathematical Morphology Operators over Concept Lattices" (Jamal)
  • 15h30 - 16h00 : "On Preference-based (soft) pattern sets" (Bruno)
  • Planning et discussion ouverte (collaborations, projets,...)

 

Lieu

Lamsade - Salle A 707.
Université Paris-Dauphine
Place du Maréchal de Lattre de Tassigny
75775 Paris Cedex 16
Métro : ligne 2 (Nation / Porte Dauphine) , station Porte Dauphine, sortie Avenue Bugeaud
RER : ligne C, station Avenue Foch
 

Demi-journée du 25 juin 2015 (Lyon)

  • 14h : "Découverte de sous-groupes dans un contexte multi-classes et avec des classes déséquilibrées" (Guillaume Bosc, LIRIS)
  • 15h : Discussion et perspectives

 

Lieu

LIRIS CNRS 5205 - INSA de Lyon
Bâtiment Blaise Pascal
3ème étage - Salle du département informatique - 502.321
69621 Villeurbanne Cedex
Tramway: lignes 1 et 4 - Arrêt "Gaston Berger"
 

Journée du 7 mai 2015 (Paris)

  • 9h45 : introduction au projet. Diapos.
  • tour de table : apport et attente de chaque partenaire sur le projet
  • la découverte instantanée de motifs. Voir les articles de Mario Boley et al. Direct Local Pattern Sampling by Efficient Two-Step Random Procedures (KDD 2011) et de Matthijs van Leeuwen Interactive Data Exploration using Pattern Mining (chapitre de livre 2014)
    • le point de vue du LIRIS (Marc) : à partir de l’article de Matthijs van Leeuwen. Diapos.
    • le point de vue du LI (Arnaud S.) : à partir de l’article de Mario Boley et al. Diapos.
  • les préférences
    • le point de vue du LORIA (Miguel). Articles sur les intégrales de Sugeno.
    • le point de vue du LI (Arnaud S.). Diapos.
  • dépôt d'une proposition d'atelier dans le cadre du GDR Madics

 

Lieu

Laboratoire LIP 6, salle 101 - couloir 26-00, 1er étage.
Le LIP6 est situé sur le site Jussieu, rotondes 24, 25 et 26.
Information d'accès
 

Missions