arXiv subjects

Victor M. Panaretos

Publications and source records attributed to Victor M. Panaretos.

46 records · Page 3Linked to original sources

arXiv2018

Procrustes Metrics on Covariance Operators and Optimal Transportation of Gaussian Processes

Covariance operators are fundamental in functional data analysis, providing the canonical means to analyse functional variation via the celebrated Karhunen--Loève expansion. These operators may themselves be subject to variation, for instance in contexts where multiple functional populations are to be compared. Statistical techniques to analyse such variation are intimately linked with the choice of metric on covariance operators, and the intrinsic infinite-dimensionality of these operators. In this paper, we describe the manifold geometry of the space of trace-class infinite-dimensional covariance operators and associated key statistical properties, under the recently proposed infinite-dimensional version of the Procrustes metric. We identify this space with that of centred Gaussian processes equipped with the Wasserstein metric of optimal transportation. The identification allows us to provide a complete description of those aspects of this manifold geometry that are important in terms of statistical inference, and establish key properties of the Fréchet mean of a random sample of covariances, as well as generative models that are canonical for such metrics and link with the problem of registration of functional data.

stat.ME↗

arXiv2017

Regression with genuinely functional errors-in-covariates

Contamination of covariates by measurement error is a classical problem in multivariate regression, where it is well known that failing to account for this contamination can result in substantial bias in the parameter estimators. The nature and degree of this effect on statistical inference is also understood to crucially depend on the specific distributional properties of the measurement error in question. When dealing with functional covariates, measurement error has thus far been modelled as additive white noise over the observation grid. Such a setting implicitly assumes that the error arises purely at the discrete sampling stage, otherwise the model can only be viewed in a weak (stochastic differential equation) sense, white noise not being a second-order process. Departing from this simple distributional setting can have serious consequences for inference, similar to the multivariate case, and current methodology will break down. In this paper, we consider the case when the additive measurement error is allowed to be a valid stochastic process. We propose a novel estimator of the slope parameter in a functional linear model, for scalar as well as functional responses, in the presence of this general measurement error specification. The proposed estimator is inspired by the multivariate regression calibration approach, but hinges on recent advances on matrix completion methods for functional data in order to handle the nontrivial (and unknown) error covariance structure. The asymptotic properties of the proposed estimators are derived. We probe the performance of the proposed estimator of slope using simulations and observe that it substantially improves upon the spectral truncation estimator based on the erroneous observations, i.e., ignoring measurement error. We also investigate the behaviour of the estimators on a real dataset on hip and knee angle curves during a gait cycle.

stat.ME↗

arXiv2016

Hybrid Regularisation of Functional Linear Models

We consider the problem of estimating the slope function in a functional regression with a scalar response and a functional covariate. This central problem of functional data analysis is well known to be ill-posed, thus requiring a regularised estimation procedure. The two most commonly used approaches are based on spectral truncation or Tikhonov regularisation of the empirical covariance operator. In principle, Tikhonov regularisation is the more canonical choice. Compared to spectral truncation, it is robust to eigenvalue ties, while it attains the optimal minimax rate of convergence in the mean squared sense, and not just in a concentration probability sense. In this paper, we show that, surprisingly, one can strictly improve upon the performance of the Tikhonov estimator in finite samples by means of a linear estimator, while retaining its stability and asymptotic properties by combining it with a form of spectral truncation. Specifically, we construct an estimator that additively decomposes the functional covariate by projecting it onto two orthogonal subspaces defined via functional PCA; it then applies Tikhonov regularisation to the one component, while leaving the other component unregularised. We prove that when the covariate is Gaussian, this hybrid estimator uniformly improves upon the MSE of the Tikhonov estimator in a non-asymptotic sense, effectively rendering it inadmissible. This domination is shown to also persist under discrete observation of the covariate function. The hybrid estimator is linear, straightforward to construct in practice, and with no computational overhead relative to the standard regularisation methods. By means of simulation, it is shown to furnish sizeable gains even for modest sample sizes.

stat.ME↗

arXiv2016

Amplitude and phase variation of point processes

We develop a canonical framework for the study of the problem of registration of multiple point processes subjected to warping, known as the problem of separation of amplitude and phase variation. The amplitude variation of a real random function $\{Y(x):x\in[0,1]\}$ corresponds to its random oscillations in the $y$-axis, typically encapsulated by its (co)variation around a mean level. In contrast, its phase variation refers to fluctuations in the $x$-axis, often caused by random time changes. We formalise similar notions for a point process, and nonparametrically separate them based on realisations of i.i.d. copies $\{Π_i\}$ of the phase-varying point process. A key element in our approach is to demonstrate that when the classical phase variation assumptions of Functional Data Analysis (FDA) are applied to the point process case, they become equivalent to conditions interpretable through the prism of the theory of optimal transportation of measure. We demonstrate that these induce a natural Wasserstein geometry tailored to the warping problem, including a formal notion of bias expressing over-registration. Within this framework, we construct nonparametric estimators that tend to avoid over-registration in finite samples. We show that they consistently estimate the warp maps, consistently estimate the structural mean, and consistently register the warped point processes, even in a sparse sampling regime. We also establish convergence rates, and derive $\sqrt{n}$-consistency and a central limit theorem in the Cox process case under dense sampling, showing rate optimality of our structural mean estimator in that case.

math.ST↗

arXiv2015

Statistical unfolding of elementary particle spectra: Empirical Bayes estimation and bias-corrected uncertainty quantification

We consider the high energy physics unfolding problem where the goal is to estimate the spectrum of elementary particles given observations distorted by the limited resolution of a particle detector. This important statistical inverse problem arising in data analysis at the Large Hadron Collider at CERN consists in estimating the intensity function of an indirectly observed Poisson point process. Unfolding typically proceeds in two steps: one first produces a regularized point estimate of the unknown intensity and then uses the variability of this estimator to form frequentist confidence intervals that quantify the uncertainty of the solution. In this paper, we propose forming the point estimate using empirical Bayes estimation which enables a data-driven choice of the regularization strength through marginal maximum likelihood estimation. Observing that neither Bayesian credible intervals nor standard bootstrap confidence intervals succeed in achieving good frequentist coverage in this problem due to the inherent bias of the regularized point estimate, we introduce an iteratively bias-corrected bootstrap technique for constructing improved confidence intervals. We show using simulations that this enables us to achieve nearly nominal frequentist coverage with only a modest increase in interval length. The proposed methodology is applied to unfolding the $Z$ boson invariant mass spectrum as measured in the CMS experiment at the Large Hadron Collider.

stat.AP↗

arXiv2014

Empirical Bayes unfolding of elementary particle spectra at the Large Hadron Collider

We consider the so-called unfolding problem in experimental high energy physics, where the goal is to estimate the true spectrum of elementary particles given observations distorted by measurement error due to the limited resolution of a particle detector. This an important statistical inverse problem arising in the analysis of data at the Large Hadron Collider at CERN. Mathematically, the problem is formalized as one of estimating the intensity function of an indirectly observed Poisson point process. Particle physicists are particularly keen on unfolding methods that feature a principled way of choosing the regularization strength and allow for the quantification of the uncertainty inherent in the solution. Though there are many approaches that have been considered by experimental physicists, it can be argued that few -- if any -- of these deal with these two key issues in a satisfactory manner. In this paper, we propose to attack the unfolding problem within the framework of empirical Bayes estimation: we consider Bayes estimators of the coefficients of a basis expansion of the unknown intensity, using a regularizing prior; and employ a Monte Carlo expectation-maximization algorithm to find the marginal maximum likelihood estimate of the hyperparameter controlling the strength of the regularization. Due to the data-driven choice of the hyperparameter, credible intervals derived using the empirical Bayes posterior lose their subjective Bayesian interpretation. Since the properties and meaning of such intervals are poorly understood, we explore instead the use of bootstrap resampling for constructing purely frequentist confidence bands for the true intensity. The performance of the proposed methodology is demonstrated using both simulations and real data from the Large Hadron Collider.

stat.AP↗

arXiv2013

Fourier analysis of stationary time series in function space

We develop the basic building blocks of a frequency domain framework for drawing statistical inferences on the second-order structure of a stationary sequence of functional data. The key element in such a context is the spectral density operator, which generalises the notion of a spectral density matrix to the functional setting, and characterises the second-order dynamics of the process. Our main tool is the functional Discrete Fourier Transform (fDFT). We derive an asymptotic Gaussian representation of the fDFT, thus allowing the transformation of the original collection of dependent random functions into a collection of approximately independent complex-valued Gaussian random functions. Our results are then employed in order to construct estimators of the spectral density operator based on smoothed versions of the periodogram kernel, the functional generalisation of the periodogram matrix. The consistency and asymptotic law of these estimators are studied in detail. As immediate consequences, we obtain central limit theorems for the mean and the long-run covariance operator of a stationary functional time series. Our results do not depend on structural modelling assumptions, but only functional versions of classical cumulant mixing conditions, and are shown to be stable under discrete observation of the individual curves.

math.ST↗

arXiv2012

Sparse approximations of protein structure from noisy random projections

Single-particle electron microscopy is a modern technique that biophysicists employ to learn the structure of proteins. It yields data that consist of noisy random projections of the protein structure in random directions, with the added complication that the projection angles cannot be observed. In order to reconstruct a three-dimensional model, the projection directions need to be estimated by use of an ad-hoc starting estimate of the unknown particle. In this paper we propose a methodology that does not rely on knowledge of the projection angles, to construct an objective data-dependent low-resolution approximation of the unknown structure that can serve as such a starting estimate. The approach assumes that the protein admits a suitable sparse representation, and employs discrete $L^1$-regularization (LASSO) as well as notions from shape theory to tackle the peculiar challenges involved in the associated inverse problem. We illustrate the approach by application to the reconstruction of an E. coli protein component called the Klenow fragment.

stat.AP↗

arXiv2012

A Conversation with David R. Brillinger

David Ross Brillinger was born on the 27th of October 1937, in Toronto, Canada. In 1955, he entered the University of Toronto, graduating with a B.A. with Honours in Pure Mathematics in 1959, while also serving as a Lieutenant in the Royal Canadian Naval Reserve. He was one of the five winners of the Putnam mathematical competition in 1958. He then went on to obtain his M.A. and Ph.D. in Mathematics at Princeton University, in 1960 and 1961, the latter under the guidance of John W. Tukey. During the period 1962--1964 he held halftime appointments as a Lecturer in Mathematics at Princeton, and a Member of Technical Staff at Bell Telephone Laboratories, Murray Hill, New Jersey. In 1964, he was appointed Lecturer and, two years later, Reader in Statistics at the London School of Economics. After spending a sabbatical year at Berkeley in 1967--1968, he returned to become Professor of Statistics in 1970, and has been there ever since. During his 40 years (and counting) as a faculty member at Berkeley, he has supervised 40 doctoral theses. He has a record of academic and professional service and has received a number of honors and awards.

stat.ME↗

arXiv2009

On random tomography with unobservable projection angles

We formulate and investigate a statistical inverse problem of a random tomographic nature, where a probability density function on $\mathbb{R}^3$ is to be recovered from observation of finitely many of its two-dimensional projections in random and unobservable directions. Such a problem is distinct from the classic problem of tomography where both the projections and the unit vectors normal to the projection plane are observable. The problem arises in single particle electron microscopy, a powerful method that biophysicists employ to learn the structure of biological macromolecules. Strictly speaking, the problem is unidentifiable and an appropriate reformulation is suggested hinging on ideas from Kendall's theory of shape. Within this setup, we demonstrate that a consistent solution to the problem may be derived, without attempting to estimate the unknown angles, if the density is assumed to admit a mixture representation.

math.ST↗