Laboratoire d’Informatique et Systèmes

Given the recent surge in developments of deep learning, this paper provides a review of the state-of-the-art deep learning techniques for audio signal processing. Speech, music, and environmental sound processing are considered side-by-side, in order to point out similarities and differences between the domains, highlighting general methods, problems, key references, and potential for cross fertilization between areas. The dominant feature representations (in particular, log-mel spectra and raw waveform) and deep learning models are reviewed, including convolutional neural networks, variants of the long short-term memory architecture, as well as more audio-specific neural network models. Subsequently, prominent deep learning application areas are covered, i.e., audio recognition (automatic speech recognition, music information retrieval, environmental sound detection, localization and tracking) and synthesis and transformation (source separation, audio enhancement, generative models for speech, sound, and music synthesis). Finally, key issues and future questions regarding deep learning applied to audio signal processing are identified.

In this brief, the optimally pruned extreme learning machine (OP-ELM) methodology is presented. It is based on the original extreme learning machine (ELM) algorithm with additional steps to make it more robust and generic. The whole methodology is presented in detail and then applied to several regression and classification problems. Results for both computational time and accuracy (mean square error) are compared to the original ELM and to three other widely used methodologies: multilayer perceptron (MLP), support vector machine (SVM), and Gaussian process (GP). As the experiments for both regression and classification illustrate, the proposed OP-ELM methodology performs several orders of magnitude faster than the other algorithms used in this brief, except the original ELM. Despite the simplicity and fast performance, the OP-ELM is still able to maintain an accuracy that is comparable to the performance of the SVM. A toolbox for the OP-ELM is publicly available online.

A keyword query is the representation of the information need of a user, and is the result of a complex cognitive process which often results in under-specification. We propose an unsupervised method namely Latent Concept Modeling (LCM) for mining and modeling latent search concepts in order to recreate the conceptual view of the original information need. We use Latent Dirichlet Allocation (LDA) to exhibit highly-specific query-related topics from pseudo-relevant feedback documents. We define these topics as the latent concepts of the user query. We perform a thorough evaluation of our approach over two large ad-hoc TREC collections. Our findings reveal that the proposed method accurately models latent concepts, while being very effective in a query expansion retrieval setting.

We address the problem of separation of mutually independent sources in nonlinear mixtures. First, we propose theoretical results and prove that in the general case, it is not possible to separate the sources without nonlinear distortion. Therefore, we focus our work on specific nonlinear mixtures known as post-nonlinear mixtures. These mixtures constituted by a linear instantaneous mixture (linear memoryless channel) followed by an unknown and invertible memoryless nonlinear distortion, are realistic models in many situations and emphasize interesting properties i.e., in such nonlinear mixtures, sources can be estimated with the same indeterminacies as in instantaneous linear mixtures. The separation structure of nonlinear mixtures is a two-stage system, namely, a nonlinear stage followed by a linear stage, the parameters of which are updated to minimize an output independence criterion expressed as a mutual information criterion. The minimization of this criterion requires knowledge or estimation of source densities or of their log-derivatives. A first algorithm based on a Gram-Charlier expansion of densities is proposed. Unfortunately, it fails for hard nonlinear mixtures. A second algorithm based on an adaptive estimation of the log-derivative of densities leads to very good performance, even with hard nonlinearities. Experiments are proposed to illustrate these results.

While in the continuous case, statistical models of histogram equalization/specification would yield exact results, their discrete counterparts fail. This is due to the fact that the cumulative distribution functions one deals with are not exactly invertible. Otherwise stated, exact histogram specification for discrete images is an ill-posed problem. Invertible cumulative distribution functions are obtained by translating the problem in a K-dimensional space and further inducing a strict ordering among image pixels. The proposed ordering refines the natural one. Experimental results and statistical models of the induced ordering are presented and several applications are discussed: image enhancement, normalization, watermarking, etc.

Abstract Assessing the presence and abundance of birds is important for monitoring specific species as well as overall ecosystem health. Many birds are most readily detected by their sounds, and thus, passive acoustic monitoring is highly appropriate. Yet acoustic monitoring is often held back by practical limitations such as the need for manual configuration, reliance on example sound libraries, low accuracy, low robustness, and limited ability to generalise to novel acoustic conditions. Here, we report outcomes from a collaborative data challenge. We present new acoustic monitoring datasets, summarise the machine learning techniques proposed by challenge teams, conduct detailed performance evaluation, and discuss how such approaches to detection can be integrated into remote monitoring projects. Multiple methods were able to attain performance of around 88% area under the receiver operating characteristic (ROC) curve (AUC), much higher performance than previous general‐purpose methods. With modern machine learning, including deep learning, general‐purpose acoustic bird detection can achieve very high retrieval rates in remote monitoring data, with no manual recalibration, and no pretraining of the detector for the target species or the acoustic conditions in the target environment.

Q fever, a worldwide zoonosis caused by Coxiella burnetii, lacks clinical specificity and may present as acute or chronic disease. Because of this polymorphism, serological confirmation is necessary to assess the diagnosis. Although microimmunofluorescence is our reference technique, the cutoff titers that are currently used to make a diagnosis of active or chronic Q fever were determined years ago with limited series of patients and sera. We determined the titers of immunoglobulin G (IgG), IgM, and IgA against both phases (I and II) of Coxiella burnetii. Rheumatoid factor was removed before testing IgM and IgA. We report here the various cutoff titers and the kinetics of antibody development from 2,218 first serum samples of patients, among whom 208 suffered from acute Q fever and 53 had chronic Q fever. In active Q fever, we have defined a low cutoff (phase II IgG titer < or = 100) below which the diagnosis cannot be made and would need further confirmation and confirmed a high cutoff (phase II IgG titer > or = 200 and phase II IgM titer > or = 50) over which the diagnosis can be made. For chronic Q fever diagnosis, phase I IgA titers are not contributive despite previous works claiming their usefulness; a phase I IgG titer of > or = 800 is highly predictive (98%) and sensitive (100%). We have also studied the possibility of rejecting or evoking the diagnosis of chronic Q fever by phase II IgG and IgA titers. This method is useful when phase I testing is not available, but the sensitivity remains low (57%).

Bacterial strain typing, or identifying bacteria at the strain level, is particularly important for diagnosis, treatment, and epidemiological surveillance of bacterial infections. This is especially the case for bacteria exhibiting high levels of antibiotic resistance or virulence, and those involved in nosocomial or pandemic infections. Strain typing also has applications in studying bacterial population dynamics. Over the last two decades, molecular methods have progressively replaced phenotypic assays to type bacterial strains. In this article, we review the current bacterial genotyping methods and classify them into three main categories: (1) DNA banding pattern-based methods, which classify bacteria according to the size of fragments generated by amplification and/or enzymatic digestion of genomic DNA, (2) DNA sequencing-based methods, which study the polymorphism of DNA sequences, and (3) DNA hybridization-based methods using nucleotidic probes. We described and compared the applications of genotyping methods to the study of bacterial strain diversity. We also discussed the selection of appropriate genotyping methods and the challenges of bacterial strain typing, described the current trends of genotyping methods, and investigated the progresses allowed by the availability of genomic sequences.

Reversible contrast mapping (RCM) is a simple integer transform that applies to pairs of pixels. For some pairs of pixels, RCM is invertible, even if the least significant bits (LSBs) of the transformed pixels are lost. The data space occupied by the LSBs is suitable for data hiding. The embedded information bit-rates of the proposed spatial domain reversible watermarking scheme are close to the highest bit-rates reported so far. The scheme does not need additional data compression, and, in terms of mathematical complexity, it appears to be the lowest complexity one proposed up to now. A very fast lookup table implementation is proposed. Robustness against cropping can be ensured as well

There is an analogy between single-chip color cameras and the human visual system in that these two systems acquire only one limited wavelength sensitivity band per spatial location. We have exploited this analogy, defining a model that characterizes a one-color per spatial position image as a coding into luminance and chrominance of the corresponding three colors per spatial position image. Luminance is defined with full spatial resolution while chrominance contains subsampled opponent colors. Moreover, luminance and chrominance follow a particular arrangement in the Fourier domain, allowing for demosaicing by spatial frequency filtering. This model shows that visual artifacts after demosaicing are due to aliasing between luminance and chrominance and could be solved using a preprocessing filter. This approach also gives new insights for the representation of single-color per spatial location images and enables formal and controllable procedures to design demosaicing algorithms that perform well compared to concurrent approaches, as demonstrated by experiments.

In this paper, we present a study on the&#13;\ncharacterization and the classification of textures. This study is performed using a set of values obtained by the computation of indexes. To obtain these indexes, we&#13;\nextract a set of data with two techniques: the computation of matrices which are statistical representations of the&#13;\ntexture and the computation of "measures". These matrices and measures are subsequently used as parameters of a function bringing real or discrete values which give information about texture features. A model of texture characterization is built based on this numerical information, for example to classify textures. An application is proposed to classify cells nuclei in order to&#13;\ndiagnose patients affected by the Progeria disease.

This paper presents a method for the optimization of the power generated from a Hybrid Renewable Energy Systems (HRES) in order to achieve the load of typical house as example of load demand. Particle Swarm Optimization Technique (PSO) is used as optimization searching algorithm due to its advantages over the other techniques for reducing the Levelized Cost of Energy (LCE) with an acceptable range of the production taking in consideration the losses between production and demand sides; the problem is defined and objective function is introduced taking in consideration fitness values sensitivity in particle swarm process. The algorithm structure was built using MATLAB software.

Identifying potentially unique features of the human cerebral cortex is a first step to understanding how evolution has shaped the brain in our species. By analyzing MR images obtained from 177 humans and 73 chimpanzees, we observed a human-specific asymmetry in the superior temporal sulcus at the heart of the communication regions and which we have named the "superior temporal asymmetrical pit" (STAP). This 45-mm-long segment ventral to Heschl's gyrus is deeper in the right hemisphere than in the left in 95% of typical human subjects, from infanthood till adulthood, and is present, irrespective of handedness, language lateralization, and sex although it is greater in males than in females. The STAP also is seen in several groups of atypical subjects including persons with situs inversus, autistic spectrum disorder, Turner syndrome, and corpus callosum agenesis. It is explained in part by the larger number of sulcal interruptions in the left than in the right hemisphere. Its early presence in the infants of this study as well as in fetuses and premature infants suggests a strong genetic influence. Because this asymmetry is barely visible in chimpanzees, we recommend the STAP region during midgestation as an important phenotype to investigate asymmetrical variations of gene expression among the primate lineage. This genetic target may provide important insights regarding the evolution of the crucial cognitive abilities sustained by this sulcus in our species, namely communication and social cognition.

International audience

drop on aluminum. The presence of proteins noticeably prolonged infectivity.

This paper describes the sequence of construction of a cell nuclei classification model by the analysis, the characterization and the classification of shape and texture. We describe first the elaboration of dedicated shape indexes and second the construction of the associated classification submodel. Then we present a new method of texture characterization, based on the construction and the analysis of statistical matrices encoding the texture. The various characterization techniques developed in this paper are systematically compared to previous approaches. In particular, we paid special attention to the results obtained by a versatile classification method using a large range of descriptors dedicated to the characterization of shapes and textures. Finally, the last classifier built with our methods achieved 88% of classification out of the 94% possible.

Bacteria have evolved several secretory pathways to release proteins into the extracellular medium. In Gram-negative bacteria, the exoproteins cross a cell envelope composed of two successive hydrophobic barriers, the cytoplasmic and outer membranes. In some cases, the protein is translocated in a single step across the cell envelope, directly from the cytoplasm to the extracellular medium. In other cases, outer membrane translocation involves an extension of the signal peptide-dependent pathway for translocation across the cytoplasmic membrane via the Sec machinery. By analogy with the so-called general export pathway (GEP), this latter route, including two separate steps across the inner and the outer membrane, was designated as the general secretory pathway (GSP) and is widely conserved among Gram-negative bacteria. In their great majority, exoproteins use the main terminal branch (MTB) of the GSP, namely the Xcp machinery in Pseudomonas aeruginosa, to reach the extracellular medium. In this review, we will use the P. aeruginosa Xcp system as a basis to discuss multiple aspects of the GSP mechanism, including machinery assembly, exoprotein recognition, energy requirement and pore formation for driving through the outer membrane.

Despite its undeniable success, classical machine learning remains a resource-intensive process. Practical computational efforts for training state-of-the-art models can now only be handled by high speed computer hardware. As this trend is expected to continue, it should come as no surprise that an increasing number of machine learning researchers are investigating the possible advantages of quantum computing. The scientific literature on Quantum Machine Learning is now enormous, and a review of its current state that can be comprehended without a physics background is necessary. The objective of this study is to present a review of Quantum Machine Learning from the perspective of conventional techniques. Departing from giving a research path from fundamental quantum theory through Quantum Machine Learning algorithms from a computer scientist's perspective, we discuss a set of basic algorithms for Quantum Machine Learning, which are the fundamental components for Quantum Machine Learning algorithms. We implement the Quanvolutional Neural Networks (QNNs) on a quantum computer to recognize handwritten digits, and compare its performance to that of its classical counterpart, the Convolutional Neural Networks (CNNs). Additionally, we implement the QSVM on the breast cancer dataset and compare it to the classical SVM. Finally, we implement the Variational Quantum Classifier (VQC) and many classical classifiers on the Iris dataset to compare their accuracies.

Melanoma is considered as one of the fatal cancer in the world, this form of skin cancer may spread to other parts of the body in case that it has not been diagnosed in an early stage. Thus, the medical field has known a great evolution with the use of automated diagnosis systems that can help doctors and even normal people to determine a certain kind of disease. In this matter, we introduce a hybrid method for melanoma skin cancer detection that can be used to examine any suspicious lesion. Our proposed system rely on the prediction of three different methods: A convolutional neural network and two classical machine learning classifiers trained with a set of features describing the borders, texture and the color of a skin lesion. These methods are then combined to improve their performances using majority voting. The experiments have shown that using the three methods together, gives the highest accuracy level.

In this paper, we present quaternion matrix algebra techniques that can be used to process the eigen analysis of a color image. Applications of principal component analysis (PCA) in image processing are numerous, and the proposed tools aim to give material for color image processing, that take into account their particular nature. For this purpose, we use the quaternion model for color images and introduce the extension of two classical techniques to their quaternionic case: singular value decomposition (SVD) and Karhunen-Loeve transform (KLT). For the quaternionic version of the KLT, we also introduce the problem of eigenvalue decomposition (EVD) of a quaternion matrix. We give the properties of these quaternion tools for color images and present their behavior on natural images. We also present a method to compute the decompositions using complex matrix algebra. Finally, we start a discussion on possible applications of the proposed techniques in color images processing.