This work aims at exploiting compressive sensing paradigms in order to reduce the cost of statistical learning tasks. We first provide a reminder of compressive sensing bases and describe some statistical analysis tasks using similar ideas. Then we describe a framework to perform parameter estimation on probabilistic mixture models in a case where training data is compressed to a fixed-size representation called a sketch. We formulate the estimation as a generalized inverse problem for which we propose a greedy algorithm. We experiment this framework and algorithm on an isotropic Gaussian mixture model. This proof of concept suggests the existence of theoretical recovery guarantees for sparse objects beyond the usual vector and matrix cases. We therefore study the generalization of linear inverse problems stability results on general signal models encompassing the standard cases and the sparse mixtures of probability distributions. We ...

During the last decades, information is being gathered and processed at an explosive rate. This fact gives rise to a very important issue, that is, how to effectively and precisely describe the information content of a given source signal or an ensemble of source signals, such that it can be stored, processed or transmitted by taking into consideration the limitations and capabilities of the several digital devices. One of the fundamental principles of signal processing for decades is the Nyquist-Shannon sampling theorem, which states that the minimum number of samples needed to reconstruct a signal without error is dictated by its bandwidth. However, there are many cases in our everyday life in which sampling at the Nyquist rate results in too many data and thus, demanding an increased processing power, as well as storage requirements. A mathematical theory that emerged ...

Today’s society is characterized by an abundance of data that is generated at an unprecedented velocity. However, much of this data is immediately thrown away by compression or information extraction. In a compressed sensing (CS) setting the inherent sparsity in many datasets is exploited by avoiding the acquisition of superfluous data in the first place. We combine this technique with tensors, or multiway arrays of numerical values, which are higher-order generalizations of vectors and matrices. As the number of entries scales exponentially in the order, tensor problems are often large-scale. We show that the combination of simple, low-rank tensor decompositions with CS effectively alleviates or even breaks the so-called curse of dimensionality. After discussing the larger data fusion optimization framework for coupled and constrained tensor decompositions, we investigate three categories of CS type algorithms to deal with large-scale problems. First, ...

Advances in compression techniques, decreasing cost of storage, and high-speed transmission have facilitated the way videos are created, stored and distributed. As a consequence, videos are now being used in many applications areas. The increase in the amount of video data deployed and used in today's applications reveals not only the importance as multimedia data type, but also led to the requirement of efficient management of video data. This management paved the way for new research areas, such as indexing and retrieval of video with respect to their spatio-temporal, visual and semantic contents. This thesis presents work towards a unified framework for semi-automated video indexing and interactive retrieval. To create an efficient index, a set of representative key frames are selected which capture and encapsulate the entire video content. This is achieved by, firstly, segmenting the video into its constituent ...

The world in which we live is becoming more and more automated, exemplified by the numerous robots, or autonomous vehicles, that operate in air, on land, or in water. These robots perform a wide array of different tasks, ranging from the dangerous, such as underground mining, to the boring, such as vacuum cleaning. In common for all different robots is that they must possess a certain degree of awareness, both of themselves and of the world in which they operate. This thesis considers aspects of two research problems associated with this, more specifically the Simultaneous Localization and Mapping (SLAM) problem and the Multiple Target Tracking (MTT) problem. The SLAM problem consists of having the robot create a map of an environment and simultaneously localize itself in the same map. One way to reduce the effect of small errors that inevitably ...

Compressed sensing (CS) is a recently introduced signal acquisition framework that goes against the traditional Nyquist sampling paradigm. CS demonstrates that a sparse, or compressible, signal can be acquired using a low rate acquisition process. Since noise is always present in practical data acquisition systems, sensing and reconstruction methods are developed assuming a Gaussian (light-tailed) model for the corrupting noise. However, when the underlying signal and/or the measurements are corrupted by impulsive noise, commonly employed linear sampling operators, coupled with Gaussian-derived reconstruction algorithms, fail to recover a close approximation of the signal. This dissertation develops robust sampling and reconstruction methods for sparse signals in the presence of impulsive noise. To achieve this objective, we make use of robust statistics theory to develop appropriate methods addressing the problem of impulsive noise in CS systems. We develop a generalized Cauchy distribution (GCD) ...

Various real-life data such as time series and multi-sensor recordings can be represented by vectors and matrices, which are one-way and two-way arrays of numerical values, respectively. Valuable information can be extracted from these measured data matrices by means of matrix factorizations in a broad range of applications within signal processing, data mining, and machine learning. While matrix-based methods are powerful and well-known tools for various applications, they are limited to single-mode variations, making them ill-suited to tackle multi-way data without loss of information. Higher-order tensors are a natural extension of vectors (first order) and matrices (second order), enabling us to represent multi-way arrays of numerical values, which have become ubiquitous in signal processing and data mining applications. By leveraging the powerful utitilies offered by tensor decompositions such as compression and uniqueness properties, we can extract more information from multi-way ...

This thesis introduces new learning algorithms for Radial Basis Function (RBF) networks. RBF networks is a feed-forward two-layer neural network used for functional approximation or pattern classification applications. The proposed training algorithms are based on robust statistics. Their theoretical performance has been assessed and compared with that of classical algorithms for training RBF networks. The applications of RBF networks described in this thesis consist of simultaneously modeling moving object segmentation and optical flow estimation in image sequences and 3-D image modeling and segmentation. A Bayesian classifier model is used for the representation of the image sequence and 3-D images. This employs an energy based description of the probability functions involved. The energy functions are represented by RBF networks whose inputs are various features drawn from the images and whose outputs are objects. The hidden units embed kernel functions. Each kernel ...

In this thesis we investigate centralized and distributed variants of sparse Bayesian learning (SBL), an effective probabilistic regression method used in machine learning. Since inference in an SBL model is not tractable in closed form, approximations are needed. We focus on the variational Bayesian approximation, as opposed to others used in the literature, for three reasons: First, it is a flexible general framework for approximate Bayesian inference that estimates probability densities including point estimates as a special case. Second, it has guaranteed convergence properties. And third, it is a deterministic approximation concept that is even applicable for high dimensional problems where non-deterministic sampling methods may be prohibitive. We resolve some inconsistencies in the literature involved in other SBL approximation techniques with regard to a proper Bayesian treatment and the incorporation of a very desired property, namely scale invariance. More specifically, ...

In the early days of speech synthesis research, understanding voice production has attracted the attention of scientists with the goal of producing intelligible speech. Later, the need to produce more natural voices led researchers to use prerecorded voice databases, containing speech units, reassembled by a concatenation algorithm. With the outgrowth of computer capacities, the length of units increased, going from diphones to non-uniform units, in the so-called unit selection framework, using a strategy referred to as 'take the best, modify the least'. Today the new challenge in voice synthesis is the production of expressive speech or singing. The mainstream solution to this problem is based on the “there is no data like more data” paradigm: emotionspecific databases are recorded and emotion-specific units are segmented. In this thesis, we propose to restart the expressive speech synthesis problem, from its original voice ...

The localization and tracking of sound sources using microphone arrays is a problem that, even if it has attracted attention from the signal processing research community for decades, remains open. In recent years, deep learning models have surpassed the state-of-the-art that had been established by classic signal processing techniques, but these models still struggle with handling rooms with strong reverberations or tracking multiple sources that dynamically appear and disappear, especially when we cannot apply any criteria to classify or order them. In this thesis, we follow the ideas of the Geometric Deep Learning framework to propose new models and techniques that mean an advance of the state-of-the-art in the aforementioned scenarios. As the input of our models, we use acoustic power maps computed using the SRP-PHAT algorithm, a classic signal processing technique that allows us to estimate the acoustic energy ...

The unprecedented processing demand, posed by the explosion of big data, challenges researchers to design efficient and adaptive machine learning algorithms that do not require persistent retraining and avoid learning redundant information. Inspired from learning techniques of intelligent biological agents, identifying transferable knowledge across learning problems has been a significant research focus to improve machine learning algorithms. In this thesis, we address the challenges of knowledge transfer through embedding spaces that capture and store hierarchical knowledge. In the first part of the thesis, we focus on the problem of cross-domain knowledge transfer. We first address zero-shot image classification, where the goal is to identify images from unseen classes using semantic descriptions of these classes. We train two coupled dictionaries which align visual and semantic domains via an intermediate embedding space. We then extend this idea by training deep networks that ...

This dissertation is the combination of three manuscripts –either published in or submitted to journals– on compressive sensing of propeller noise for detection, identification and localization of water crafts. Propeller noise, as a result of rotating blades, is broadband and radiates through water dominating underwater acoustic noise spectrum especially when cavitation develops. Propeller cavitation yields cyclostationary noise which can be modeled by amplitude modulation, i.e., the envelope-carrier product. The envelope consists of the so-called propeller tonals representing propeller characteristics which is used to identify water crafts whereas the carrier is a stationary broadband process. Sampling for propeller noise processing yields large data sizes due to Nyquist rate and multiple sensor deployment. A compressive sensing scheme is proposed for efficient sampling of second-order cyclostationary propeller noise since the spectral correlation function of the amplitude modulation model is sparse as shown in ...

Next generation search engines will enable query formulations, other than text, relying on visual information encoded in terms of images and shapes. The 3D search technology, in particular, targets specialized application domains ranging from computer aided-design and manufacturing to cultural heritage archival and presentation. Content-based retrieval research aims at developing search engines that would allow users to perform a query by similarity of content. This thesis deals with two fundamentals problems in content-based 3D object retrieval: (1) How to describe a 3D shape to obtain a reliable representative for the subsequent task of similarity search? (2) How to supervise the search process to learn inter-shape similarities for more effective and semantic retrieval? Concerning the first problem, we develop a novel 3D shape description scheme based on probability density of multivariate local surface features. We constructively obtain local characterizations of 3D ...

Speech signals carry important information about a speaker such as age, gender, language, accent and emotional/psychological state. Automatic recognition of speaker characteristics has a wide range of commercial, medical and forensic applications such as interactive voice response systems, service customization, natural human-machine interaction, recognizing the type of pathology of speakers, and directing the forensic investigation process. This research aims to develop accurate methods and tools to identify different physical characteristics of the speakers. Due to the lack of required databases, among all characteristics of speakers, our experiments cover gender recognition, age estimation, language recognition and accent/dialect identification. However, similar approaches and techniques can be applied to identify other characteristics such as emotional/psychological state. For speaker characterization, we first convert variable-duration speech signals into fixed-dimensional vectors suitable for classification/regression algorithms. This is performed by fitting a probability density function to acoustic ...