Stéphan Clémençon

Tree-Based Ranking Methods

This paper investigates how recursive partitioning methods can be adapted to the bipartite ranking problem. In ranking, the pursued goal is global: based on past data, define an order on the whole input space X, so that positive instances take up the top ranks with maximum probability. The most natural way to order all instances consists of projecting the input data onto the real line through a real-valued scoring function s and use the natural order on R. The accuracy of the ordering induced by a candidate s is classically measured in terms of the ROC curve or the AUC. Here we discuss the design of tree-structured scoring functions obtained by recursively maximizing the AUC criterion. The connection with recursive piecewise linear approximation of the optimal ROC curve both in the L1-sense and in the Linfin-sense is highlighted. A novel tree-based algorithm for ranking, called TreeRank, is proposed. Consistency results and generalization bounds of functional nature are established for this ranking method, when considering either the L1 or Linfin distance. We also describe committee-based learning procedures using TreeRank as a ldquobase ranker,rdquo in order to overcome obvious drawbacks of such a top-down partitioning technique. Simulation results on artificial data are also displayed.

The HIV/AIDS epidemic in Cuba: description and tentative explanation of its low HIV prevalence

BackgroundThe Cuban HIV/AIDS epidemic has the lowest prevalence rate of the Caribbean region. The objective of this paper is to give an overview of the HIV/AIDS epidemic in Cuba and to explore the reasons for this low prevalence.MethodsData were obtained from the Cuban HIV/AIDS programme established in 1983. This programme has an extensive adult HIV testing policy, including testing of all pregnant women. HIV and AIDS cases have been recorded since 1986. Persons found to be HIV-positive are interviewed on their sexual behaviour and partners. Tracing and voluntary testing of these partners are organised. Epidemiological description of this epidemic was obtained from analysis of this data set. Using elementary mathematical analyses, we estimated the coverage of the detection system (percentage of HIV-positive adults detected) and the average period between HIV infection and detection. Estimated HIV prevalence rates were corrected to account for the coverage.ResultsHIV prevalence has increased since 1996. In 2005, the prevalence among pregnant women was 1.2 per 10,000 (16/137000). Estimated HIV prevalence among 15- to 49-year-olds was 8.1 per 10,000 (4913/6065000; 95%CI: 7.9 per 10,000 – 8.3 per 10,000). Most (77%) of the HIV-positive adults were men, most (85.1%) of the detected HIV-positive men were reported as having sex with men (MSM), and most of the HIV-positive women reported having had sex with MSM. The average period between HIV infection and detection was estimated to be 2.1 years (IQR = 1.7 – 2.2 years). We estimated that, for the year 2005, 79.6% (IQR: 77.3 – 81.4%) of the HIV-positive persons were detected.ConclusionMSM drive the HIV epidemic in Cuba. The extensive HIV testing policy may be an important factor in explaining the low HIV prevalence. To reduce the HIV epidemic in Cuba, the epidemic among MSM should be addressed. To understand this epidemic further, data on sexual behaviour should be collected. Now that antiretroviral therapy is more widely available, the Cuban policy, based on intensive HIV testing and tracing of partners, may be considered as a possible policy to control HIV/AIDS epidemics in other countries.

Ranking and scoring using empirical risk minimization

A general model is proposed for studying ranking problems. We investigate learning methods based on empirical minimization of the natural estimates of the ranking risk. The empirical estimates are of the form of a U-statistic. Inequalities from the theory of U-statistics and U-processes are used to obtain performance bounds for the empirical risk minimizers. Convex risk minimization methods are also studied to give a theoretical framework for ranking algorithms based on boosting and support vector machines. Just like in binary classification, fast rates of convergence are achieved under certain noise assumption. General sufficient conditions are proposed in several special cases that guarantee fast rates of convergence.

Efficacy of Vaccination against HPV Infections to Prevent Cervical Cancer in France: Present Assessment and Pathways to Improve Vaccination Policies

Background Seventy percent of sexually active individuals will be infected with Human Papillomavirus (HPV) during their lifetime. These infections are incriminated for almost all cervical cancers. In France, 3,068 new cases of cervical cancer and 1,067 deaths from cervical cancer occurred in 2005. Two vaccines against HPV infections are currently available and vaccination policies aim to decrease the incidence of HPV infections and of cervical cancers. In France, vaccine coverage has been reported to be low. Methods We developed a dynamic model for the heterosexual transmission of Human Papillomavirus types 16 and 18, which are covered by available vaccines. A deterministic model was used with stratification on gender, age and sexual behavior. Immunity obtained from vaccination was taken into account. The model was calibrated using French data of cervical cancer incidence. Results In view of current vaccine coverage and screening, we expected a 32% and 83% reduction in the incidence of cervical cancers due to HPV 16/18, after 20 years and 50 years of vaccine introduction respectively. Vaccine coverage and screening rates were assumed to be constant. However, increasing vaccine coverage in women or vaccinating girls before 14 showed a better impact on cervical cancer incidence. On the other hand, performing vaccination in men improves the effect on cervical cancer incidence only moderately, compared to strategies in females only. Conclusion While current vaccination policies may significantly decrease cervical cancer incidence, other supplementary strategies in females could be considered in order to improve vaccination efficacy.

New approach for the assessment of cluster diets

Dietary risk assessment is a major public health concern, positioned in the context of establishing overall food safety policy. It requires some understanding of population food choices although geographical location and social-cultural environment are variable. Several years ago, a cluster analysis based on FAO consumption data, ranging from 1990 to 1994, was at the origin of the 13, so called, GEMS/Food cluster diets. This analysis required the initial identification of 19 food markers based on geographical and cultural differences. This paper proposes a new modelling of FAO food consumption database in order to define new cluster diets based on updated consumption data from 2002 to 2007 and better adapted statistical methods. Two statistical methods were combined to extract, consumption systems that generate a substructure from the initial food consumption database and then by deriving a clustering of countries according to their consumption system profiles. The clustering resulted in 17 cluster diets composed of 2 up to 30 countries. The few discrepancies between these new clusters and former ones may be due to more recent data, and to the fact that the new approach is based on another mathematical modelling which does not require any initial identification of food markers.

Adaptive partitioning schemes for bipartite ranking

Recursive partitioning methods are among the most popular techniques in machine learning. The purpose of this paper is to investigate how to adapt this methodology to the bipartite ranking problem. Following in the footsteps of the TreeRank approach developed in Clémençon and Vayatis (Proceedings of the 2008 Conference on Algorithmic Learning Theory, 2008 and IEEE Trans. Inf. Theory 55(9):4316–4336, 2009), we present tree-structured algorithms designed for learning to rank instances based on classification data. The main contributions of the present work are the following: the practical implementation of the TreeRank algorithm, well-founded solutions to the crucial issues related to the splitting rule and the choice of the “right” size for the ranking tree. From the angle embraced in this paper, splitting is viewed as a cost-sensitive classification task with data-dependent cost. Hence, up to straightforward modifications, any classification algorithm may serve as a splitting rule. Also, we propose to implement a cost-complexity pruning method after the growing stage in order to produce a “right-sized” ranking sub-tree with large AUC. In particular, performance bounds are established for pruning schemes inspired by recent work on nonparametric model selection. Eventually, we propose indicators for variable importance and variable dependence, plus various simulation studies illustrating the potential of our method.

Sharp Bounds for the Tails of Functionals of Markov Chains

This paper is devoted to establishing sharp bounds for deviation probabilities of partial sums

Efficient eigen-updating for spectral graph clustering

\sum_{i=1}^{n}f(X_{i})

EMOTHAW: A Novel Database for Emotional State Recognition From Handwriting and Drawing

, where

Extraction of Food Consumption Systems by Nonnegative Matrix Factorization (NMF) for the Assessment of Food Choices

X=(X_{n})_{n\in{\bf N}}