Introduction to the IEEE Journal on Selected Topics in Signal Processing and IEEE Transactions on Signal and Information Processing Over Networks Joint Special Issue on Graph Signal Processing

Pier Luigi Dragotti, Pascal Frossard
2017
Article

Résumé

The papers in this special issue are intended to address some of the main research challenges in Graph Signal Processing by presenting a collection of the latest advances in the domain. These papers examine key representation, learning and processing aspects for signals living on graphs and networks, as well as new methods and applications in graph signal processing. Numerous applications rely on the processing of high dimensional data that reside on irregular or otherwise unordered structures that are naturally modeled as networks. The need for new tools to process such data has led to the emergence of the field of graph signal processing, which merges algebraic and spectral graph theoretic concepts with computational harmonic analysis to process signals on structures such as graphs. This important new paradigm in signal processing research, coupled with its numerous applications in very different domains, has fueled the rapid development of an inter-disciplinary research community that has been working on theoretical aspects of graph signal processing and applications to diverse problems such as big data analysis, coding and compression of 3D point clouds, biological data processing, and brain network analysis.

Official source

https://infoscience.epfl.ch/record/230479?ln=fr

À propos de ce résultat

Cette page est générée automatiquement et peut contenir des informations qui ne sont pas correctes, complètes, à jour ou pertinentes par rapport à votre recherche. Il en va de même pour toutes les autres pages de ce site. Veillez à vérifier les informations auprès des sources officielles de l'EPFL.

Concepts associés

Chargement

Publications associées

Chargement

Statistical limits of high-dimensional inference problems

Clément Dominique Luneau

This thesis focuses on two kinds of statistical inference problems in signal processing and data science. The first problem is the estimation of a structured informative tensor from the observation of a noisy tensor in which it is buried. The structure comes from the possibility to decompose the informative tensor as the sum of a small number of rank-one tensors (small compared to its size). Such structure has applications in data science where data, organized into arrays, can often be explained by the interaction between a few features characteristic of the problem. The second problem is the estimation of a signal input to a feedforward neural network whose output is observed. It is relevant for many applications (phase retrieval, quantized signals) where the relation between the measurements and the quantities of interest is not linear. We look at these two statistical models in different high-dimensional limits corresponding to situations where the amount of observations and size of the signal become infinitely large. These asymptotic regimes are motivated by the ever-increasing computational power and storage capacity that make possible the processing and handling of large data sets. We take an information-theoretic approach in order to establish fundamental statistical limits of estimation in high-dimensional regimes. In particular, the main contributions of this thesis are the proofs of exact formulas for the asymptotic normalized mutual information associated with these inference problems. These are low-dimensional variational formulas that can nonetheless capture the behavior of a large system where each component interacts with all the others. Owing to the relationship between mutual information and Bayesian inference, we use the solutions to these variational problems to rigorously predict the asymptotic minimum mean-square error (MMSE), the error achieved by the (Bayes) optimal estimator. We can thus compare algorithmic performances to the statistical limit given by the MMSE. Variational formulas for the mutual information are referred to as replica symmetric (RS) ansätze due to the predictions of the heuristic replica method from statistical physics. In the past decade proofs of the validity of these predictions started to emerge. The general strategy is to show that the RS ansatz is both an upper and lower bound on the asymptotic normalized mutual information. The present work leverages on the adaptive interpolation method that proposes a unified way to prove the two bounds. We extend the adaptive interpolation to situations where the order parameter of the problem is not a scalar but a matrix, and to high-dimensional regimes that differ from the one for which the RS formula is usually conjectured. Our proofs also demonstrate the modularity of the method. Indeed, using statistical models previously studied in the literature as building blocks of more complex ones (e.g., estimated signal with a model of structure), we derive RS formulas for the normalized mutual information associated with estimation problems that are relevant to modern applications.

EPFL2021

Chargement

Wavelet footprints and frames for signal processing and communication

Pier Luigi Dragotti

Linear transforms and expansions are fundamental mathematical tools of signal processing. In particular, the wavelet transform has played an important role in several signal processing tasks, compression being a prime example. A signal can be represented in a basis or a frame. Frames, which are an extension of bases to overcomplete sets of vectors, are sometimes preferred to bases because of their greater design freedom and recently they have had an important impact in signal processing. This thesis focuses on the design of new bases and frames for signal processing and communications. The first contribution of the thesis is the exploration of the use of oversampled filter banks, which represent a possible way to implement frames, to robust communications. Normally, transforms are used as a pure source coding method and a reliable communication is obtained with a channel coder which follows the source coder. In this case, we use frames to achieve efficient source compression and robustness to transmission errors at the same time. In the context of pure source transform coding, we investigate the performance of the wavelet transform and study the dependency of the wavelet coefficients across scales. This analysis leads to the design of a new expansion which provides an efficient representation of piecewise smooth signals. We call footprints the elements of this expansion. The main property of footprints is that they efficiently characterize the singular structures of the signal, which usually carry the main information. We show that algorithms based on footprints outperform wavelet methods in different applications such as denoising, compression and deconvolution. Finally, we study a particular source coding technique called multiple description coding. This technique is used for data transmission over unreliable networks. In multiple description coding, the coder generates several different descriptions of the same signal and the decoder can produce a useful reconstruction of the source with any received subset of these descriptions. We study the problem of multiple description coding of stationary Gaussian sources with memory. First, we compute the multiple description rate distortion region for these sources. Then, we develop an algorithm for the design of optimal critically sampled filter banks for multiple description coding of Gaussian sources.

EPFL2002

Chargement

"Random spikes" belong to the common language used by engineers, physicists and biologists to describe events associated with time records, locations in space, or more generally, space-time events. Indeed, data and signals consisting of, or structured by, sequences of events are omnipresent in communications, biology, computer science and signal processing. Relevant examples can be found in traffic intensity and neurobiological data, pulse-coded transmission, and sampling. This thesis is concerned by random spike fields and by the complex signals described as the result of various operations on the basic event stream or spike field, such as filtering, jittering, delaying, thinning, clustering, sampling and modulating. More precisely, complex signals are obtained in a modular way by adding specific features to a basic model. This modular approach greatly simplifies the computations and allows to treat highly complex model such as the ones occurring in ultra-wide bandwidth or multipath transmissions. We present a systematic study of the properties of random spikes and related complex signals. More specifically, we focus on second order properties, which are conveniently represented by the spectrum of the signal. These properties are particularly attractive and play an important role in signal analysis. Indeed, they are relatively accessible and yet they provide important informations. Our first contribution is theoretical. As well as presenting a modular approach for the construction of complex signals, we derive formulas for the computation of the spectrum that preserve such modularity: each additional feature added to a basic model appear as a separate and explicit contribution in the corresponding basic spectrum. Moreover, these formula are very general. For instance, the basic point process is not assumed to be a homogeneous Poisson process but it can be any second order stationary process with a given spectrum. In summary, they provide very useful tools for model analysis. We then give applications of the theoretical results: spectral formulas for traffic analysis, pulse based signals used in spread spectrum communications, and randomly sampled signal.

EPFL2005

Chargement

Wavelet footprints and frames for signal processing and communication

Pier Luigi Dragotti

EPFL2002

Statistical limits of high-dimensional inference problems

Clément Dominique Luneau

EPFL2021

EPFL2005