IDENTIFICATION

Pierre Lafaye de Micheaux

Date of birth: March 27, 1973. Marital status: married with two children. Citizenships : Canada, France, Switzerland.

DEGREES

ACADEMIC POSITIONS

INDUSTRIAL EXPERIENCE

INTERNATIONAL IMPACT

Short research periods in other departments

Invited conferences¹

Sherbrooke, Canada (2010). Hyderabad, India (2009). Montpellier, France (2004).

Short courses and tutorials

Grants and scholarships

Academic distinction

Elective responsabilities

Administrative responsabilities

Reviewer for international journals

Bernoulli, Canadian Journal of Statistics, Computational Statistics and Data Analysis, Journal of Multivariate Analysis, Journal of Statistical Planning and Inference, Journal of Statistical Software, Mathematical Reviews, Medical Image Computing and Computer Assisted Intervention (MICCAI) Proceedings.

PUBLICATIONS AND SOFTWARES

Research interests

Asymptotics, Biostatistics, Bootstrap, Complex random fields, Hypothesis testing theory, Independent Component Analysis, Linear regression methods, Model selection, Multivariate statistics, Neuroscience, Stochastic processes, Teaching, Time series analysis.

Published (or accepted) articles in international refereed journals

1. 2011: Tabelow K., Clayden J.D., Lafaye de Micheaux P., Polzehl J., Schmid V.J., Whitcher B., Image Analysis and Statistical Inference in Neuroimaging with R. NeuroImage, 9 pages.
2. 2010: Duchesne P. and Lafaye de Micheaux P., Computing the distribution of quadratic forms: Further comparisons between the Liu-Tan-Zhang approximation and exact methods. Computational Statistics and Data Analysis, 54(4), p. 858-862, April.
3. 2009: Coeurjolly J.F., Drouilhet R., Lafaye de Micheaux P. and Robineau J.F., asympTest: A simple R package for performing classical parametric statistical tests and confidence intervals in large samples. The R Journal, 1(2), p. 26-30, December.
4. 2009: Lafaye de Micheaux P., Liquet B., ConvergenceConcepts: an R package to investigate various modes of convergence. The R Journal, 1(2), p. 18-25, December.
5. 2009: Bilodeau M. and Lafaye de Micheaux P., A-dependence statistics for mutual and serial independence of categorical variables. Journal of Statistical Planning and Inference, 139(7), 1 July, p. 2407-2419.
6. 2009: Lafaye de Micheaux P. and Liquet B., Understanding Convergence Concepts: a Graphical Simulation Based Approach. The American Statistician, 63(2), p. 173-178.
7. 2008: Cenier T., Amat C., Litaudon P., Garcia S., Lafaye de Micheaux P., Liquet B., Roux S. and Buonviso N., Odor vapor pressure and quality modulate local field potential oscillatory patterns in the olfactory bulb of the anesthetized rat. European Journal of Neuroscience, 28(6), March, p. 1432-1440.
8. 2007: Dubois M., Lafaye de Micheaux P., Noël M.P. and Valdois S., Pre-orthographical constraints on visual word recognition : Evidence from a case study of developmental surface dyslexia. Cognitive Neuropsychology, 24(6), p. 623-660.
9. 2007: Beran R., Bilodeau M. and Lafaye de Micheaux P., Nonparametric tests of independence between random vectors. Journal of Multivariate Analysis, 98(9), p. 1805-1824.
10. 2005: Bilodeau M. and Lafaye de Micheaux P., A multivariate empirical characteristic function test of independence with normal marginals. Journal of Multivariate Analysis, 95(2), p. 345-369.
11. 2004: Ducharme G.R. and Lafaye de Micheaux P., Goodness-of-fit tests of normality for the innovations in ARMA models. Journal of Time Series Analysis, 25(3), p. 373-395.

Books

1. 201?: Lafaye de Micheaux Pierre, Drouilhet Rémy, Liquet Benoit, The R software, Mastering the language - Performing statistical analysis, Springer - Collection: Computing and statistics, in progress.
2. 2010: Lafaye de Micheaux Pierre, Drouilhet Rémy, Liquet Benoit, Le logiciel R, Maîtriser le langage - Effectuer des analyses statistiques, Springer - Collection: Statistique et probabilités appliquées, Vol. 1, 1st Edition., XVI, 530 p., Broché, ISBN: 978-2-8178-0114-8. (French)

Articles submitted to international refereed journals

1. 2010: Desgagné A., Lafaye de Micheaux P. and Leblanc A., Goodness-of-fit test of normality against generalized exponential power alternatives, 18 pages.
2. 2010: Bordier C., Dojat, M. and Lafaye de Micheaux P., Temporal and Spatial Independent Component Analysis for fMRI data sets embedded in a R package, 23 pages.

International and national communications, proceedings²

1. 2010: Ducharme G., Lafaye de Micheaux P., Marchina B., Tests d’ajustement basés sur la fonction caractéristique, Colloque de statistique Sherbrooke-Montpellier, Sherbrooke, Canada, October 6-7.
2. 2010: Tabelow K., Clayden J.D., Lafaye de Micheaux P., Polzehl J., Schmid V.J., Whitcher B., Image Analysis and Statistical Inference in Neuroimaging with R, 16th Annual Meeting of the Organization for Human Brain Mapping, Barcelona, Spain, June 6-10. Poster.
3. 2010: Lafaye de Micheaux P., Liquet B., Le Package R ConvergenceConcepts: Un nouvel outil graphique pour l’étude de quelques modes de convergence de variables aléatoires, 42èmes Journées de Statistique, Marseille, France.
4. 2009: Lafaye de Micheaux P., The R-Package ConvergenceConcepts: A new graphical tool to investigate various modes of convergence, International Conference on Frontiers of Interface Between Statistics and Sciences, Conference Proceedings, Hyderabad, India, December 31. Invited speaker.
5. 2009: Bordier C., Dojat M. and Lafaye de Micheaux P., AnalyzeFMRI: an R package to perform statistical analysis on fMRI, The R User Conference, Rennes, France.
6. 2009: Bordier C., Dojat M. and Lafaye de Micheaux P., L’Analyse en Composantes Indépendantes Temporelle en Imagerie par Résonance Magnétique Fonctionnelle, 41ème Congrès de la SFDS, Bordeaux, France: http://hal.inria.fr/inria-00386659. And also Chairman of session “Tests”.
7. 2008: Bilodeau M., Lafaye de Micheaux P., A-dependence statistics for mutual and serial independence of categorical variables, First conference of the Moroccan society of Applied Mathematics, École nationale de l’Industrie Minérale, Rabat, Morocco.
8. 2008: Desgagné A., Lafaye de Micheaux P., Leblanc A., Goodness-of-fit test of normality against GEP alternatives, 36^th conference of the Statistical Society of Canada Meeting, Ottawa, Canada.
9. 2008: Bilodeau M., Lafaye de Micheaux P., A-dependence statistics for mutual and serial independence of categorical variables, 36^th conference of the Statistical Society of Canada Meeting, Ottawa, Canada.
10. 2008: Lafaye de Micheaux P., Multi- and univariate Neyman smooth test of normality in a time series context, 36^th conference of the Statistical Society of Canada Meeting, Ottawa, Canada.
11. 2007: Desjardins S., Desgagné A., Lafaye de Micheaux P., Liquet B., Generalization of the Paired T-Test for the Missing Values Case, Joint Statistical Meetings, Salt Lake City, USA. Poster.
12. 2006: Beran R., Bilodeau M., Lafaye de Micheaux P., Nonparametric tests of independence between random vectors, Swiss Statistics Meeting, Lugano, Switzerland.
13. 2006: Beran R., Bilodeau M., Lafaye de Micheaux P., Nonparametric tests of independence between random vectors, XV^th Int. Symposium on Mathematical Methods Applied to the Sciences, San Jose, Costa Rica.
14. 2005: Desgagné A., Lafaye de Micheaux P., Leblanc A., Test de normalité contre des alternatives de type GEP, 37ème Congrès de la SFDS, Pau, France.
15. 2004: Lafaye de Micheaux P., Léger C., Régression linéaire adaptative multiple: une approche bootstrap, 32^th conference of the Statistical Society of Canada Meeting, Montréal, Canada.
16. 2004: Lafaye de Micheaux P., Léger C., Régression linéaire adaptative multiple: une approche bootstrap, 36ème Congrès de la SFDS, Montpellier, France. Invited Speaker in Plenary Session.
17. 2003: Bilodeau M., Lafaye de Micheaux P., A multivariate empirical characteristic function test of independence with normal marginals, Swiss Statistics Meeting, Montreux, Switzerland.
18. 2003: Desgagné A., Lafaye de Micheaux P., Leblanc A., Test de normalité basé sur la famille des densités GEP, Journées de l’optimisation 2003, Montréal, Canada.

Collaborative work that should be achieved in a near future

1. Lafaye de Micheaux P. and Léger C., Data-driven multiple linear regression: a bootstrap approach.
2. Lafaye de Micheaux P., Neyman smooth goodness-of-fit test for normality in ARMA and VARMA models.
3. Desgagné A., Lafaye de Micheaux P. and Leblanc A., P-kurtosis and goodness-of-fit test for normality.

Collaborative work already started

1. Gimenez O., Lafaye de Micheaux P., Bootstrapping χ² goodness-of-fit test statistics in sparse one way multinomials: an empirical study.
2. Desgagné A., Lafaye de Micheaux P. and Leblanc A., A new test of double exponentiality based on the P-kurtosis.
3. Lafaye de Micheaux P., Marchina B., On the effect of parameter estimation in limiting χ² U- and V-statistics involving complex-valued components.
4. Lafaye de Micheaux P., Vinod H. D., Approximate comparison of convergence rates for consistent estimators in finite samples.

Technical reports and thesis

1. Tabelow K., Clayden J.D., Lafaye de Micheaux P., Polzehl J., Schmid V.J., Whitcher B. (Décembre 2010). Image Analysis and Statistical Inference in Neuroimaging with R, Technical report 1578, Weierstrass Institute for Applied Analysis and Stochastics, 9 pages.
2. Bordier C., Dojat, M. and Lafaye de Micheaux P. (Décembre 2010). Temporal and Spatial Independent Component Analysis for fMRI data sets embedded in a R package, arXiv 1012.0269v1, 23 pages.
3. Coeurjolly J.-F., Drouilhet, R., Lafaye de Micheaux P. and Robineau, J.-F. (February 2009). asympTest: an R package for performing parametric statistical tests and confidence intervals based on the central limit theorem. Technical report hal-00358375. Laboratoire Jean Kuntzmann. Université de Grenoble, 18 pages.
4. Lafaye de Micheaux P. Multivariate Statistical Methods in FMRI. (June 2007). Master’s thesis. Institut National Polytechnique de Grenoble. (In french).
5. Lafaye de Micheaux P. Independence tests in multivariate analysis and normality tests in ARMA models. (December 2002). Ph.D. thesis. Université de Montréal and Université Montpellier II.
6. Ducharme G.R. and Lafaye de Micheaux P. (February 2002). Goodness-of-fit tests of normality for the innovations in ARMA models. Technical report number 02-02. Groupe de biostatistique et d’analyse des systèmes. Université Montpellier II, 34 pages.

Computing and software experience

1. Developper of AnalyzeFMRI R package (initiated by Marchini, J.), available on CRAN. Functions for I/O FMRI data in various formats and for treating brain images with ICA methods. http://cran.r-project.org/package=AnalyzeFMRI
2. Developper of ConvergenceConcepts R package (with Liquet, B.) to check and visualize convergence of a sequence of random variables in law, in probability, in r-th mean and almost surely (see [I4] and [I6]).
   http://cran.r-project.org/package=ConvergenceConcepts
3. Creation of R package asympTest, partly described in [I3] and available on CRAN, with Coeurjolly J.F., Drouilhet R. and Robineau J.F. http://cran.r-project.org/package=asympTest
4. Creation of R package CompQuadForm, available on CRAN, described in [I2] (with Duchesne P.). http://cran.r-project.org/package=CompQuadForm

TALKS

Invited seminars in Canada

Invited seminars in France

Invited seminars in India

Université de Montréal’s graduate students seminar

PRESENT AND FUTURE RESEARCH ACTIVITIES

My research work is done in the Department of Mathematics and Statistics of Université de Montréal (Canada), and also in the Grenoble Institute of Neuroscience (France).

My mid-term objectives are to pursue the development of tools in the multivariate time series domain, and to apply these tools to study the temporal data acquired during functional magnetic resonance imaging (FMRI) or electro-encephalogram (EEG) experiments. I also plan to keep on working on Independent Component Analysis methods because these tools are well adapted to study these neurological data. I plan to keep on working on the development of innovative statistical hypothesis testing procedures and also on their potential applications to clinical trials.

Research projects in the Department of Mathematics and Statistics (DMS)

• Independence tests: I am interested in the development of some specific dependence measures to construct non parametric independence tests. We have proposed, with Bilodeau M. (Pr. Université de Montréal) an interesting characterization of independence using the Moëbius function. This has permitted us to obtain a mutual independence test between the p multivariate marginals (supposed Gaussian) of a random vector. This test is based on some Cramér-von Mises functionnal of a certain complex-valued random field involving characteristic functions. We also show how to test the serial independence of random vectors. This work has been published ([I10]) and has been presented in a conference ([C17]). Next, we have extended this work, with Beran R. (Pr. University of California, Davis), to the case where the marginals are supposed unknown and are estimated by their empirical distribution function. The limit law of our statistic depends upon the unkown marginals laws and we have used Bootstrap methods, and proved their consistency, to obtain critical values. This work has also been published ([I9]) and presented in two conferences ([C12], [C13]). In both cases, the test statistics obtained have as a limiting distribution an infinite sum of chi-square weighted independent random variables. We proposed a methodology ([I2]) and an R package ([P4]) to obtain the critical values of such a distribution.
We have adapted these results with Bilodeau M. to generalize of the χ² test of independence to several qualitative variables. A paper describing this work has been published ([I5]) and presented in two conferences ([C7] and [C9]). An R package allowing to perform all the tests previously described is under construction.

• Goodness-of-fit tests (independent or dependent data):
- Goodness-of-fit tests to a distribution L(λ) which are based on a Cramér-von Mises functionnal, involving a distance measure between the empirical characteristic function of the sample and the postulated theoretical characteristic function under the null, often lead to U- or V-statistics T_n(λ) with a χ²-type limiting distribution. The parameter λ being unkown, one has to replace it by some estimator λ. A natural question that arises is then to evaluate the impact of this substitution on the limiting distribution of T_n(λ). We have been working on this problem with a master student in the case where T_n(λ) involves complex-valued components. We have obtained some rather general conditions describing this impact. An interesting follow-up to this work will be to construct goodness-of-fit tests for many distributions ([E3]).

- The theory of Neyman’s smooth goodness-of-fit tests consists in embedding the density of the observations in a parametric family vast enough to take into account the way the data could depart from the postulated null hypothesis. Next Rao’s test strategy, which consists in testing the nullity of some parameters of the embedding family, is applied in order to obtain a test statistic which has an asymptotic χ² distribution under the null. This technique continues to receive interest as shown in the following papers: Bickel et al. (Ann. Statist., 34(2), 721-741, (2006)) or Langovoy (arXiv:0708.0169, submitted to Ann. Statist., 2007). Following this approach, we have constructed with Ducharme G. (Pr. Montpellier II) a goodness-of-fit test of the normality of the innovations of an ARMA model with known mean. This work has been published ([R6] and [I11]). I began in 2007 to extend the results exposed in [I11] to the case of an unknown mean and to multivariate VARMA processes. Note also that the choice of the integer determining the size of the exponential embedding family is data-driven using BIC like criteria, and that the computation of the statistical test quantiles is refined through a third order correction. The results obtained so far have been presented in a conference ([C10]). They are also collected in a paper that should be submitted soon ([A2]).

- In the same vein, a collaboration started with Desgagné A. (Pr. UQAM, Montréal) and Leblanc A. (Pr. Université du Manitoba) has led us to propose a Gaussianity test by using as an embedding family the set of the Generalized Exponetial Power densities introduced by Angers (Metron 58, 81-108, (2000)). This family of distributions enables one to characterize and order the tails of a vast class of densities. The test we propose is thus well adapted to detect departures from the normality in the tails of the distribution. We have presented this work in some conferences ([C8], [C14] and [C18]) and also in a recently submitted paper ([S1]). This collaboration is going on at the moment (see [A3], [E2]).

• Bootstrap methods: A research work started with Léger C. (Pr. Université de Montréal) is directed towards the construction of confidence intervals taking into account the uncertainty introduced by a preliminary selection of models. We obtained theoretical results on the convergence (in probability and almost surely) of the Bootstrap methods used in this work. A part of this work has already been presented in some conferences ([C15], [C16]). A paper is in preparation ([A1]).
Another previous work ([E1]), started with Gimenez O. (Researcher CNRS, Centre d’Écologie Fonctionnelle et Évolutive, Montpellier), consists in using Bootstrap methods for testing goodness-of-fit of the multinomial distribution in the context of sparse data, a situation which often occurs in capture-recapture experiments.

• Teaching and softwares: I am interested in the teaching of statistical concepts by using and producing computer tools. Firstly, we attempt to propose an intuitive explanation of the four classical modes of convergence of a random sequence (see the articles [I6] and [I4], and the R package ConvergenceConcepts ([P2]). Secondly, we aim at providing students with an R package enabling them to perform an asymptotic version (robust to non normality) of the classical parametric tests they learn in undergraduate courses (mean, variance and proportion tests, for one or two samples), see [I3] and [P3].

Research projects in the Grenoble Institute of Neuroscience (GIN)

The functional magnetic resonance imaging (fMRI) is a non-invasive technique which enables one to study cognitive processes on ill or healthy patients. An fMRI experiment usually consists in presenting the subject with stimuli related to the cognitive task to be explored. A set of volumic images of the brain in activity is then acquired during these periods. The resulting datasets are huge (around 500,000×200). The aim is then to determine which parts of the cortex are activated by the stimuli, and also how these various areas interact. My activity within the GIN come in two complementary parts: on the one hand, I am bringing a biostatistical support to neuroscience researchers in order to help them using adequately some already existing statistical procedures; on the other hand, I participate to the development of theoretical tools to answer the new questions they ask themselves.

• Independent Component Analysis: I am trying to apply Independent Component Analysis (ICA) methods on these data sets, and to study how to avoid some difficulties encountered when using these methods in an fMRI context. We have to select and order the spatial or temporal components obtained, check the independence hypothesis of the source images and monitor the injection of a priori knowledge. A paper ([S2]), co-authored with Dojat M. (INSERM Researcher), describing how to perform a temporal ICA (supposed to be impracticable because of the huge size of the data) on fMRI data has been submitted. This work uses properties of the Singluar Value Decomposition and has been presented in some Neuroscience seminars. It is implemented in [P1].
• MoDyC project: With Coeurjolly J.-F. (MCF, Grenoble 2) and Achard S., we have recently formed a small team of statisticians with the objective to explore some theoretical and methodological avenues like ICA, construction of graphs of brain functionnal connectivity or also caracterizing cogitive schemes by means of Hurst exponents. The aim of the MoDyC project, conducted in collaboration with Sato M. (Researcher CNRS, GIPSA-LAB - Department Speech & Cognition), is to obtain a spatial and temporal dynamical description of the motor, somato-sensor and auditive brain activations, linked to the production and perception of vowels from the french language. This one year project has been funded from the Institut National Polytechnique de Grenoble.

ÉTHEMATIC MOBILITY

TEACHING AND SUPERVISION EXPERIENCE

Supervision of graduate students

• Students in Ph.D.

• Students in M.Sc.

Supervision of undergraduate students

Teaching

INVOLVEMENT IN THE ACADEMIC COMMUNITY

MEMBERSHIP IN PROFESSIONAL SOCIETIES

Statistical Society of Canada, Swiss Statistical Society, French Statistical Society.

LANGUAGES