Séance du 7 avril 2014

Séance organisée par Estelle Kuhn et Mathilde Mougeot.

Lieu : IHP, Amphithéâtre Hermite.

14h00 : Ron S. Kenett (KPA Ltd., Raanana, Israel University of Turin & Center for Risk Engineering, NYU)

Titre : Quantitative and Qualitative Aspects of Bayesian Networks: A General Approach for Integrating Expert Opinions and Structured Data

Résumé : Modern data consists of textual unstructured data, social networks mapping relationships, subjective information based on expert opinions and quantitative data collected in small or very large amounts. Combining data from various such sources has been a major challenge for statisticians in a wide range of application areas. Bayesian networks build on graphical analysis and Bayesian inference to represent descriptive causality maps that combine variables of different types (Kenett, 2013a). The talk will present such applications of Bayesian networks to web site usability (Harel et al, Kenett et al, 2009) operational risks (Kenett and Raanan, 2010), customer satisfaction surveys (Kenett and Salini, 2011), biotechnology and healthcare systems (Kenett, 2012) and the testing of web services (Bai and Kenett, 2012). In particular the talk will present a methodology based on risk assessment of scenarios applied in the RISCOSS project on risks in open source software adoption (www.riscoss.eu). We conclude with a general perspective of the role of applied statistics in the context of various developments like data science and statistical engineering (Kenett, 2013b).

References

1. Bai, X., Kenett, R.S. and Yu, W. (2012). Risk Assessment and Adaptive Group Testing of Semantic Web Services. International Journal of Software Engineering and Knowledge Engineering, 2012.

2. Harel. A., Kenett, R.S. and Ruggeri, F. (2009). Modeling Web Usability Diagnostics on the basis of Usage Statistics, in Statistical Methods in eCommerce Research, W. Jank and G. Shmueli (editors), Wiley.

3. Kenett, R.S. and Raanan, Y. (2010). Operational Risk Management: a practical approach to intelligent data analysis, Wiley and Sons. http://www.wiley.com/WileyCDA/WileyTitle/productCd-047074748X.html

4. Kenett R.S. and Salini S. (2011). Modern Analysis of Customer Satisfaction Surveys: with applications using R, John Wiley and Sons, Chichester: UK.

5. Kenett, R.S. (2012). Risk Analysis in Drug Manufacturing and Healthcare, in Statistical Methods in Healthcare, Faltin, F., Kenett, R.S. and Ruggeri, F. (editors in chief), John Wiley and Sons.

6. Kenett, R.S. (2013a). Applications of Bayesian Networks. http://ssrn.com/abstract=2172713

7. Kenett, R.S. (2013b). Statistics: A Life Cycle View. http://ssrn.com/abstract=2315556

15h00 : Olivier Catoni (CNRS - ENS - INRIA CLASSIC)

Titre : Markov substitute models and statistical inference in linguistics.

Résumé : In this talk, we will present a new model of random .nite sequences of words. It can be seen as a generalization of Markov chains based

on more general conditional independence assumptions. The Markov substitute assumption is related to the reversibility of some conditional probability

kernels that substitute randomly in a sentence an expression with another one. We will present tests and parameter estimators based on PAC-Bayes

theorems. We will also discuss the links between this model and toric grammars, that we introduced in a previous work as a possible stochastic language

model. Although our research is focussed on computational linguistics, we think that Markov substitute models could be useful in other frameworks

where extensions of the Markov model are desirable and where typically hidden Markov models are used.

(joint work with Thomas Mainguy).

16h00 : Dominique Picard (Université Paris Diderot)

Titre : Procedures bayésiennes adaptatives pour des données à structure géometrique

Résumé : Nous nous posons le problème d'étendre les résultats de Ghosal, Ghosh et van der Vaart, sur la vitesse de concentration des mesures a posteriori, au cas de des données ayant une structure géométrique. Ce problème, illustré par l'estimation de densité pour des données sur la sphère ou sur d'autres espaces structurés comme des variétés riemanniennes, des graphes, ou des arbres, pose des problèmes intéressants qui demandent d'analyser et d'utiliser la nature géométrique de l'espace considéré. Nous considérerons des mesures a priori gaussiennes. Le problème en particulier de l'adaptation montre la nécessité de relier la loi a priori à l'analyse harmonique de la structure considérée. Avant de traiter l'aspect géométrique, nous envisagerons le problème sous l'angle plus simple d'une mesure a priori gaussienne sur des coefficients d'ondelettes.