We derive a central limit theorem for triangular arrays of possibly nonstationary random variables satisfying a condition of weak dependence in the sense of Doukhan and Louhichi [Stoch. Proc. Appl. 84 (1999) 313–342]. The proof uses a new variant of the Lindeberg method: the behavior of the partial sums is compared to that of partial sums of dependent Gaussian random variables. We also discuss a few applications in statistics which show that our central limit theorem is tailor-made for statistics of different type.

We consider strictly stationary infinitely divisible processes and first extend the mixing conditions given in Maruyama [Theory Probab. Appl. 15 (1970) 1–22] and Rosiński and Żak [Stoc. Proc. Appl. 61 (1996) 277–288] from the univariate to the d-dimensional case. Thereafter, we show that multivariate Lévy-driven mixed moving average processes satisfy these conditions and hence a wide range of well-known processes such as superpositions of Ornstein − Uhlenbeck (supOU) processes or (fractionally integrated) continuous time autoregressive moving average (CARMA) processes are always mixing. Finally, mixing of the log-returns and the integrated volatility process of a multivariate supOU type stochastic volatility model, recently introduced in Barndorff − Nielsen and Stelzer [Math. Finance 23 (2013) 275–296], is established.

We consider the problem of estimating the integral of the square of a densityf from the observation of a n sample. Our method to estimate $\int_{\mathbb{R}} f^2(x){\rm d}x$ isbased on model selection via some penalized criterion. We prove that our estimator achieves the adaptive rates established by Efroimovich and Low on classes of smooth functions. A key point of the proof is an exponentialinequality for U-statistics of order 2 due to Houdré and Reynaud.

We elucidate the asymptotics of the Ls -quantization error induced by a sequence of Lr -optimal n-quantizers of aprobability distribution P on $\mathbb{R}^d$ when s > r. In particular we show that under natural assumptions, the optimal rate is preserved aslong as s < r+d (and for everys in the case of a compactly supported distribution). We derive some applications of these results to the error bounds for quantization based cubatureformulae in numerical integration on $\mathbb{R}^d$ and on the Wiener space.

This paper deals with the likelihood ratio test (LRT) for testing hypotheseson the mixing measure in mixture models with or without structural parameter. The main result gives the asymptotic distribution of the LRTstatisticsunder some conditions that are proved to be almost necessary.A detailed solution is given for two testing problems: thetest of a single distribution against any mixture, with application to Gaussian, Poisson andbinomial distributions; the test of the number of populations in afinite mixture with or without structural parameter.

The work developed in the paper concerns the multivariate fractional Brownian motion (mfBm) viewed through the lens of the wavelet transform. After recalling some basic properties on the mfBm, we calculate the correlation structure of its wavelet transform. We particularly study the asymptotic behaviour of the correlation, showing that if the analyzing wavelet has a sufficient number of null first order moments, the decomposition eliminates any possible long-range (inter)dependence. The cross-spectral density is also considered in a second part. Its existence is proved and its evaluation is performed using a von Bahr–Essen like representation of the function sign(t)|t|α. The behaviour of the cross-spectral density of the wavelet field at the zero frequency is also developed and confirms the results provided by the asymptotic analysis of the correlation.

In this paper, we analyze the celebrated EM algorithm from the point of view of proximal point algorithms. More precisely, we study a new type of generalization of the EM procedure introduced in [Chretien and Hero (1998)] and called Kullback-proximal algorithms. The proximal framework allows us to prove new results concerning the cluster points. An essential contribution is a detailed analysis of the case where some cluster points lie on the boundary of the parameter space.

Assessing the number of clusters of a statistical population is one of the essential issues of unsupervised learning. Given n independent observations X1,...,Xn drawn from an unknown multivariate probability density f, we propose a new approach to estimate the number of connected components, or clusters, of the t-level set $\mathcal L(t)=\{x:f(x) \geq t\}$ . The basic idea is to form a rough skeleton of the set $\mathcal L(t)$ using any preliminary estimator of f, and to count the number of connected components of the resulting graph. Under mild analytic conditions on f, and using tools from differential geometry, we establish the consistency of our method.

We study the scenario of graph-based clustering algorithms such as spectral clustering. Given a set of data points, one first has to construct a graph on the data points and then apply a graph clustering algorithm to find a suitable partition of the graph. Our main question is if and how the construction of the graph (choice of the graph, choice of parameters, choice of weights) influences the outcome of the final clustering result. To this end we study the convergence of cluster quality measures such as the normalized cut or the Cheeger cut on various kinds of random geometric graphs as the sample size tends to infinity. It turns out that the limit values of the same objective function are systematically different on different types of graphs. This implies that clustering results systematically depend on the graph and can be very different for different types of graph. We provide examples to illustrate the implications on spectral clustering.

In many practical situations sample sizes are not sufficiently largeand estimators based on such samples may not be satisfactory interms of their variances. At the same time it is not unusual thatsome auxiliary information about the parameters of interest isavailable. This paper considers a method of using auxiliaryinformation for improving properties of the estimators based on acurrent sample only. In particular, it is assumed that theinformation is available as a number of estimates based on samplesobtained from some other mutually independent data sources. Thismethod uses the fact that there is a correlation effect betweenestimators based on the current sample and auxiliary informationfrom other sources. If variance covariance matrices of vectors ofestimators used in the estimating procedure are known, this methodproduces more efficient estimates in terms of their variancescompared to the estimates based on the current sample only. If thesevariance-covariance matrices are not known, their consistentestimates can be used as well such that the large sample propertiesof the method remain unchangeable. This approach allows to improvestatistical properties of many standard estimators such as anempirical cumulative distribution function, empirical characteristicfunction, and Nelson-Aalen cumulative hazard estimator.

We consider an estimate of the mode θ of a multivariate probability density f with support in $\mathbb R^d$ using a kernel estimate f n drawn from a sample X1,...,Xn . The estimate θn is defined as any x in {X1,...,Xn } such that $f_n(x)=\max_{i=1, \hdots,n} f_n(X_i)$ . It is shown that θn behaves asymptotically as any maximizer ${\hat{\theta}}_n$ of f n . More precisely, we prove that for any sequence $(r_n)_{n\geq 1}$ of positive real numbers such that $r_n\to\infty$ and $r_n^d\log n/n\to 0$ , one has $r_n\,\|\theta_n-{\hat{\theta}}_n\| \to 0$ in probability. The asymptotic normality of θn follows without further work.

We prove the local asymptotic normality for the full parameters of the normal inverseGaussian Lévy process X, when we observe high-frequency data XΔn,X2Δn,...,XnΔnwith sampling mesh Δn → 0 and the terminalsampling time nΔn → ∞. Therate of convergence turns out to be (√nΔn, √nΔn, √n, √n) for the dominating parameter(α,β,δ,μ), where α stands for the heaviness of thetails, β the degree of skewness, δ the scale, andμ the location. The essential feature in our study is that the suitablynormalized increments of X in small time is approximatelyCauchy-distributed, which specifically comes out in the form of the asymptotic Fisherinformation matrix.

The estimation of probabilistic deformable template models incomputer vision or of probabilistic atlases in Computational Anatomyare core issues in both fields.A first coherent statistical framework where the geometrical variability ismodelled as a hiddenrandom variable has been given by [S. Allassonnière et al., J. Roy. Stat. Soc.69 (2007) 3–29]. They introduce a Bayesian approach and mixture of them to estimate deformable template models.A consistent stochastic algorithm has been introduced in [S. Allassonnière et al. (in revision)] to face the problem encountered in [S. Allassonnière et al., J. Roy. Stat. Soc.69 (2007) 3–29] for theconvergence of the estimation algorithm for the one component model inthe presence of noise.We propose here to go on in this direction of using some “SAEM-like”algorithm to approximate the MAP estimator in the general Bayesian setting ofmixture of deformable template models.We also prove the convergence of our algorithm toward a criticalpoint of the penalised likelihood of the observations andillustrate this with handwritten digit images and medical images.

Let Y be a Ornstein–Uhlenbeck diffusion governed by astationary and ergodic process X : dYt = a(Xt)Yt dt + σ(Xt)dWt,Y0 = y0 . We establish that under the condition α = Eµ(a(X0)) < 0 with μ the stationary distribution of the regime process X, the diffusion Y is ergodic. We also consider conditions for the existence of moments for theinvariant law of Y when X is a Markov jump process having a finite number of states.Using results on random difference equationson one hand and the fact that conditionally toX, Y is Gaussian on the other hand, we give such a condition for the existence of the moment of order s ≥ 0. Actually we recover in this case a result that Basak et al. [J. Math. Anal. Appl. 202 (1996) 604–622] have established using the theory of stochastic control of linear systems.

We estimate the anisotropic index of an anisotropic fractional Brownian field. For all directions, we give a convergent estimator of the value of the anisotropic index in this direction, based on generalized quadratic variations. We also prove a central limit theorem. First we present a result of identification that relies on the asymptotic behavior of the spectral density of a process. Then, we define Radon transforms of the anisotropic fractional Brownian field and prove that these processes admit a spectral density satisfying the previous assumptions. Finally we use simulated fields to test the proposed estimator in different anisotropic and isotropic cases. Results show that the estimator behaves similarly in all cases and is able to detect anisotropy quiteaccurately.

We present a method for estimating the edge of a two-dimensionalbounded set, given a finite random set of points drawn from the interior.The estimator is based both on a Parzen-Rosenblatt kernel and extreme values of point processes. We give conditionsfor various kinds of convergence and asymptotic normality.We propose a method of reducing the negative bias and edge effects,illustrated by some simulations.

This paper deals with the problem of estimating the level setsL(c) =  {F(x) ≥ c},with c ∈ (0,1), of an unknown distribution function F on ℝ+2. A plug-inapproach is followed. That is, given a consistent estimatorFn of F, we estimateL(c) byLn(c) =  {Fn(x) ≥ c}.In our setting, non-compactness property is a priori required for thelevel sets to estimate. We state consistency results with respect to the Hausdorffdistance and the volume of the symmetric difference. Our results are motivated byapplications in multivariate risk theory. In particular we propose a new bivariate versionof the conditional tail expectation by conditioning the two-dimensional random vector tobe in the level set L(c). We also present simulated andreal examples which illustrate our theoretical results.

We discuss various properties of Probabilistic Cellular Automata, suchas the structure of the set of stationary measures and multiplicity ofstationary measures (or phase transition) for reversible models.

We recall necessary notions about the geometry and harmonic analysis on a hyperbolic space and provide lecture notes about homogeneous random functions parameterized by this space. The general principles are illustrated by construction of numerous examples analogous to Euclidean case. We also give a brief survey of the fields parameterized by Euclidean spheres. At the end we give a list of important open questions in hyperbolic case.

We consider stationary processes with long memory which are non-Gaussian and representedas Hermite polynomials of a Gaussian process. We focus on the corresponding waveletcoefficients and study the asymptotic behavior of the sum of their squares since this sumis often used for estimating the long–memory parameter. We show that the limit is notGaussian but can be expressed using the non-Gaussian Rosenblatt process defined as aWiener–Itô integral of order 2. This happens even if the original process is definedthrough a Hermite polynomial of order higher than 2.