Searcharxiv⌕ Search

arXiv subjects

Stéphan Clémençon

Publications and source records attributed to Stéphan Clémençon.

44 records · Page 3Linked to original sources
arXiv2012

An Empirical Comparison of V-fold Penalisation and Cross Validation for Model Selection in Distribution-Free Regression

Model selection is a crucial issue in machine-learning and a wide variety of penalisation methods (with possibly data dependent complexity penalties) have recently been introduced for this purpose. However their empirical performance is generally not well documented in the literature. It is the goal of this paper to investigate to which extent such recent techniques can be successfully used for the tuning of both the regularisation and kernel parameters in support vector regression (SVR) and the complexity measure in regression trees (CART). This task is traditionally solved via V-fold cross-validation (VFCV), which gives efficient results for a reasonable computational cost. A disadvantage however of VFCV is that the procedure is known to provide an asymptotically suboptimal risk estimate as the number of examples tends to infinity. Recently, a penalisation procedure called V-fold penalisation has been proposed to improve on VFCV, supported by theoretical arguments. Here we report on an extensive set of experiments comparing V-fold penalisation and VFCV for SVR/CART calibration on several benchmark datasets. We highlight cases in which VFCV and V-fold penalisation provide poor estimates of the risk respectively and introduce a modified penalisation technique to reduce the estimation error.

stat.ML↗
arXiv2012

Dissemination of Health Information within Social Networks

In this paper, we investigate, how information about a common food born health hazard, known as Campylobacter, spreads once it was delivered to a random sample of individuals in France. The central question addressed here is how individual characteristics and the various aspects of social network influence the spread of information. A key claim of our paper is that information diffusion processes occur in a patterned network of social ties of heterogeneous actors. Our percolation models show that the characteristics of the recipients of the information matter as much if not more than the characteristics of the sender of the information in deciding whether the information will be transmitted through a particular tie. We also found that at least for this particular advisory, it is not the perceived need of the recipients for the information that matters but their general interest in the topic.

cs.SI↗
arXiv2012

Hierarchical clustering for graph visualization

This paper describes a graph visualization methodology based on hierarchical maximal modularity clustering, with interactive and significant coarsening and refining possibilities. An application of this method to HIV epidemic analysis in Cuba is outlined.

stat.AP↗
arXiv2012

Visual Mining of Epidemic Networks

We show how an interactive graph visualization method based on maximal modularity clustering can be used to explore a large epidemic network. The visual representation is used to display statistical tests results that expose the relations between the propagation of HIV in a sexual contact network and the sexual orientation of the patients.

stat.AP↗
arXiv2008

A stochastic SIR model with contact-tracing: large population limits and statistical inference

A stochastic epidemic model accounting for the effect of contact-tracing on the spread of an infectious disease is studied. Precisely, individuals identified as infected may contribute to detecting other infectious individuals by providing information related to persons with whom they have had possibly infectious contacts. The population evolves through demographic, infection and detection processes, in a way that its temporal evolution is described by a stochastic Markov process, of which the component accounting for the contact-tracing feature is assumed to be valued in a space of point measures. For adequate scalings of the demographic, infection and detection rates, it is shown to converge to the weak deterministic solution of a PDE system, as a parameter n, interpreted as the population size roughly speaking, becomes large. From the perspective of the analysis of infectious disease data, this approximation result may serve as a key tool for exploring the asymptotic properties of standard inference methods such as maximum likelihood estimation. We state preliminary statistical results in this context. Eventually, relation of the model to the available data of the HIV epidemic in Cuba, in which country a contact-tracing detection system has been set up since 1986, is investigated and numerical applications are carried out.

math.PR↗
arXiv2007

Ranking the best instances

We formulate the local ranking problem in the framework of bipartite ranking where the goal is to focus on the best instances. We propose a methodology based on the construction of real-valued scoring functions. We study empirical risk minimization of dedicated statistics which involve empirical quantiles of the scores. We first state the problem of finding the best instances which can be cast as a classification problem with mass constraint. Next, we develop special performance measures for the local ranking problem which extend the Area Under an ROC Curve (AUC/AROC) criterion and describe the optimal elements of these new criteria. We also highlight the fact that the goal of ranking the best instances cannot be achieved in a stage-wise manner where first, the best instances would be tentatively identified and then a standard AUC criterion could be applied. Eventually, we state preliminary statistical results for the local ranking problem.

math.ST↗
arXiv2006

Ranking and empirical minimization of U-statistics

The problem of ranking/ordering instances, instead of simply classifying them, has recently gained much attention in machine learning. In this paper we formulate the ranking problem in a rigorous statistical framework. The goal is to learn a ranking rule for deciding, among two instances, which one is "better," with minimum ranking risk. Since the natural estimates of the risk are of the form of a U-statistic, results of the theory of U-processes are required for investigating the consistency of empirical risk minimizers. We establish in particular a tail inequality for degenerate U-processes, and apply it for showing that fast rates of convergence may be achieved under specific noise assumptions, just like in classification. Convex risk minimization methods are also studied.

math.ST↗