École de Technologie Supérieure

This paper presents an extension of our previous work which proposes a new speaker representation for speaker verification. In this modeling, a new low-dimensional speaker- and channel-dependent space is defined using a simple factor analysis. This space is named the total variability space because it models both speaker and channel variabilities. Two speaker verification systems are proposed which use this new representation. The first system is a support vector machine-based system that uses the cosine kernel to estimate the similarity between the input data. The second system directly uses the cosine similarity as the final decision score. We tested three channel compensation techniques in the total variability space, which are within-class covariance normalization (WCCN), linear discriminate analysis (LDA), and nuisance attribute projection (NAP). We found that the best results are obtained when LDA is followed by WCCN. We achieved an equal error rate (EER) of 1.12% and MinDCF of 0.0094 using the cosine distance scoring on the male English trials of the core condition of the NIST 2008 Speaker Recognition Evaluation dataset. We also obtained 4% absolute EER improvement for both-gender trials on the 10 s-10 s condition compared to the classical joint factor analysis scoring.

Active filtering of electric power has now become a mature technology for harmonic and reactive power compensation in two-wire (single phase), three-wire (three phase without neutral), and four-wire (three phase with neutral) AC power networks with nonlinear loads. This paper presents a comprehensive review of active filter (AF) configurations, control strategies, selection of components, other related economic and technical considerations, and their selection for specific applications. It is aimed at providing a broad perspective on the status of AF technology to researchers and application engineers dealing with power quality issues. A list of more than 200 research publications on the subject is also appended for a quick reference.

In recent years, free space optical (FSO) communication has gained significant importance owing to its unique features: large bandwidth, license free spectrum, high data rate, easy and quick deployability, less power, and low mass requirements. FSO communication uses optical carrier in the near infrared band to establish either terrestrial links within the Earth's atmosphere or inter-satellite/deep space links or ground-to-satellite/satellite-to-ground links. It also finds its applications in remote sensing, radio astronomy, military, disaster recovery, last mile access, backhaul for wireless cellular networks, and many more. However, despite of great potential of FSO communication, its performance is limited by the adverse effects (viz., absorption, scattering, and turbulence) of the atmospheric channel. Out of these three effects, the atmospheric turbulence is a major challenge that may lead to serious degradation in the bit error rate performance of the system and make the communication link infeasible. This paper presents a comprehensive survey on various challenges faced by FSO communication system for ground-to-satellite/satellite-to-ground and inter-satellite links. It also provides details of various performance mitigation techniques in order to have high link availability and reliability. The first part of this paper will focus on various types of impairments that pose a serious challenge to the performance of optical communication system for ground-to-satellite/satellite-to-ground and inter-satellite links. The latter part of this paper will provide the reader with an exhaustive review of various techniques both at physical layer as well as at the other layers (link, network, or transport layer) to combat the adverse effects of the atmosphere. It also uniquely presents a recently developed technique using orbital angular momentum for utilizing the high capacity advantage of optical carrier in case of space-based and near-Earth optical communication links. This survey provides the reader with comprehensive details on the use of space-based optical backhaul links in order to provide high capacity and low cost backhaul solutions.

Three-phase AC-DC converters have been developed to a matured level with improved power quality in terms of power-factor correction, reduced total harmonic distortion at input AC mains, and regulated DC output in buck, boost, buck-boost, multilevel, and multipulse modes with unidirectional and bidirectional power flow. This paper presents an exhaustive review of three-phase improved power quality AC-DC converters (IPQCs) configurations, control strategies, selection of components, comparative factors, recent trends, their suitability, and selection for specific applications. It is aimed at presenting a state of the art on the IPQC technology to researchers, designers, and application engineers dealing with three-phase AC-DC converters. A classified list of around 450 research articles on IPQCs is also appended for a quick reference.

Underwater wireless information transfer is of great interest to the military, industry, and the scientific community, as it plays an important role in tactical surveillance, pollution monitoring, oil control and maintenance, offshore explorations, climate change monitoring, and oceanography research. In order to facilitate all these activities, there is an increase in the number of unmanned vehicles or devices deployed underwater, which require high bandwidth and high capacity for information transfer underwater. Although tremendous progress has been made in the field of acoustic communication underwater, however, it is limited by bandwidth. All this has led to the proliferation of underwater optical wireless communication (UOWC), as it provides higher data rates than the traditional acoustic communication systems with significantly lower power consumption and simpler computational complexities for short-range wireless links. UOWC has many potential applications ranging from deep oceans to coastal waters. However, the biggest challenge for underwater wireless communication originates from the fundamental characteristics of ocean or sea water; addressing these challenges requires a thorough understanding of complex physio-chemical biological systems. In this paper, the main focus is to understand the feasibility and the reliability of high data rate underwater optical links due to various propagation phenomena that impact the performance of the system. This paper provides an exhaustive overview of recent advances in UOWC. Channel characterization, modulation schemes, coding techniques, and various sources of noise which are specific to UOWC are discussed. This paper not only provides exhaustive research in underwater optical communication but also aims to provide the development of new ideas that would help in the growth of future underwater communication. A hybrid approach to an acousto-optic communication system is presented that complements the existing acoustic system, resulting in high data rates, low latency, and an energy-efficient system.

This paper presents an easy-to-use battery model applied to dynamic simulation software. The simulation model uses only the battery State-Of-Charge (SOC) as a state variable in order to avoid the algebraic loop problem. It is shown that this model, composed of a controlled voltage source in series with a resistance, can accurately represent four types of battery chemistries. The model's parameters can easily be extracted from the manufacturer's discharge curve, which allows for an easy use of the model. A method is described to extract the model's parameters and to approximate the internal resistance. The model is validated by superimposing the results with the manufacturer's discharge curves. Finally, the battery model is included in the SimPowerSystems (SPS) simulation software and is used in the Hybrid Electric Vehicle (HEV) demo. The results for the battery and for the DC-DC converter are analysed and they show that the model can accurately represent the general behaviour of the battery.

This paper presents an improved and easy-to-use battery dynamic model. The charge and the discharge dynamics of the battery model are validated experimentally with four batteries types. An interesting feature of this model is the simplicity to extract the dynamic model parameters from batteries datasheets. Only three points on the manufacturer’s discharge curve in steady state are required to obtain the parameters. Finally, the battery model is included in the SimPowerSystems simulation software and used in a detailed simulation of an electric vehicle based on a hybrid fuel cell-battery power source. The results show that the model can accurately represent the dynamic behaviour of the battery.

Multiple-input multiple output (MIMO) communication architecture has recently emerged as a new paradigm for wireless communications in rich multipath environment, which has spectral efficiencies far beyond those offered by conventional techniques. The channel capacity of the MIMO architecture in independent Rayleigh channels scales linearly as the number of antennas. However, the correlation of a real-world wireless channel may result in a substantial degradation of the MIMO architecture performance. In this letter, we investigate the MIMO channel capacity in correlated channels using the exponential correlation matrix model. We prove that, for this model, an increase in correlation is equivalent to a decrease in signal-to-noise ratio (SNR). For example, r=0.7 is the same as 3-dB decrease in SNR.

Background: With the development of smart grids, accurate electric load forecasting has become increasingly important as it can help power companies in better load scheduling and reduce excessive electricity production. However, developing and selecting accurate time series models is a challenging task as this requires training several different models for selecting the best amongst them along with substantial feature engineering to derive informative features and finding optimal time lags, a commonly used input features for time series models. Methods: Our approach uses machine learning and a long short-term memory (LSTM)-based neural network with various configurations to construct forecasting models for short to medium term aggregate load forecasting. The research solves above mentioned problems by training several linear and non-linear machine learning algorithms and picking the best as baseline, choosing best features using wrapper and embedded feature selection methods and finally using genetic algorithm (GA) to find optimal time lags and number of layers for LSTM model predictive performance optimization. Results: Using France metropolitan’s electricity consumption data as a case study, obtained results show that LSTM based model has shown high accuracy then machine learning model that is optimized with hyperparameter tuning. Using the best features, optimal lags, layers and training various LSTM configurations further improved forecasting accuracy. Conclusions: A LSTM model using only optimally selected time lagged features captured all the characteristics of complex time series and showed decreased Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) for medium to long range forecasting for a wider metropolitan area.

The security issue impacting the Internet-of-Things (IoT) paradigm has recently attracted significant attention from the research community. To this end, several surveys were put forward addressing various IoT-centric topics, including intrusion detection systems, threat modeling, and emerging technologies. In contrast, in this paper, we exclusively focus on the ever-evolving IoT vulnerabilities. In this context, we initially provide a comprehensive classification of state-of-the-art surveys, which address various dimensions of the IoT paradigm. This aims at facilitating IoT research endeavors by amalgamating, comparing, and contrasting dispersed research contributions. Subsequently, we provide a unique taxonomy, which sheds the light on IoT vulnerabilities, their attack vectors, impacts on numerous security objectives, attacks which exploit such vulnerabilities, corresponding remediation methodologies and currently offered operational cyber security capabilities to infer and monitor such weaknesses. This aims at providing the reader with a multidimensional research perspective related to IoT vulnerabilities, including their technical details and consequences, which is postulated to be leveraged for remediation objectives. Additionally, motivated by the lack of empirical (and malicious) data related to the IoT paradigm, this paper also presents a first look on Internet-scale IoT exploitations by drawing upon more than 1.2 GB of macroscopic, passive measurements' data. This aims at practically highlighting the severity of the IoT problem, while providing operational situational awareness capabilities, which undoubtedly would aid in the mitigation task, at large. Insightful findings, inferences and outcomes in addition to open challenges and research problems are also disclosed in this paper, which we hope would pave the way for future research endeavors addressing theoretical and empirical aspects related to the imperative topic of IoT security.

Driven by privacy concerns and the visions of deep learning, the last four years have witnessed a paradigm shift in the applicability mechanism of machine learning (ML). An emerging model, called federated learning (FL), is rising above both centralized systems and on-site analysis, to be a new fashioned design for ML implementation. It is a privacy-preserving decentralized approach, which keeps raw data on devices and involves local ML training while eliminating data communication overhead. A federation of the learned and shared models is then performed on a central server to aggregate and share the built knowledge among participants. This article starts by examining and comparing different ML-based deployment architectures, followed by in-depth and in-breadth investigation on FL. Compared to the existing reviews in the field, we provide in this survey a new classification of FL topics and research fields based on thorough analysis of the main technical challenges and current related work. In this context, we elaborate comprehensive taxonomies covering various challenging aspects, contributions, and trends in the literature, including core system models and designs, application areas, privacy and security, and resource management. Furthermore, we discuss important challenges and open research directions toward more robust FL systems.

As healthcare systems around the world face challenges such as escalating costs, limited access, and growing demand for personalized care, artificial intelligence (AI) is emerging as a key force for transformation. This review is motivated by the urgent need to harness AI's potential to mitigate these issues and aims to critically assess AI's integration in different healthcare domains. We explore how AI empowers clinical decision-making, optimizes hospital operation and management, refines medical image analysis, and revolutionizes patient care and monitoring through AI-powered wearables. Through several case studies, we review how AI has transformed specific healthcare domains and discuss the remaining challenges and possible solutions. Additionally, we will discuss methodologies for assessing AI healthcare solutions, ethical challenges of AI deployment, and the importance of data privacy and bias mitigation for responsible technology use. By presenting a critical assessment of AI's transformative potential, this review equips researchers with a deeper understanding of AI's current and future impact on healthcare. It encourages an interdisciplinary dialogue between researchers, clinicians, and technologists to navigate the complexities of AI implementation, fostering the development of AI-driven solutions that prioritize ethical standards, equity, and a patient-centered approach.

Abstract. The European Centre for Medium-Range Weather Forecasts (ECMWF) recently released its most advanced reanalysis product, the ERA5 dataset. It was designed and generated with methods giving it multiple advantages over the previous release, the ERA-Interim reanalysis product. Notably, it has a finer spatial resolution, is archived at the hourly time step, uses a more advanced assimilation system and includes more sources of data. This paper aims to evaluate the ERA5 reanalysis as a potential reference dataset for hydrological modelling by considering the ERA5 precipitation and temperatures as proxies for observations in the hydrological modelling process, using two lumped hydrological models over 3138 North American catchments. This study shows that ERA5-based hydrological modelling performance is equivalent to using observations over most of North America, with the exception of the eastern half of the US, where observations lead to consistently better performance. ERA5 temperature and precipitation biases are consistently reduced compared to ERA-Interim and systematically more accurate for hydrological modelling. Differences between ERA5, ERA-Interim and observation datasets are mostly linked to precipitation, as temperature only marginally influences the hydrological simulation outcomes.

The integration of nanotechnology into three-dimensional printing (3DP) offers huge potential and opportunities for the manufacturing of 3D engineered materials exhibiting optimized properties and multifunctionality. The literature relating to different 3DP techniques used to fabricate 3D structures at the macro- and microscale made of nanocomposite materials is reviewed here. The current state-of-the-art fabrication methods, their main characteristics (e.g., resolutions, advantages, limitations), the process parameters, and materials requirements are discussed. A comprehensive review is carried out on the use of metal- and carbon-based nanomaterials incorporated into polymers or hydrogels for the manufacturing of 3D structures, mostly at the microscale, using different 3D-printing techniques. Several methods, including but not limited to micro-stereolithography, extrusion-based direct-write technologies, inkjet-printing techniques, and popular powder-bed technology, are discussed. Various examples of 3D nanocomposite macro- and microstructures manufactured using different 3D-printing technologies for a wide range of domains such as microelectromechanical systems (MEMS), lab-on-a-chip, microfluidics, engineered materials and composites, microelectronics, tissue engineering, and biosystems are reviewed. Parallel advances on materials and techniques are still required in order to employ the full potential of 3D printing of multifunctional nanocomposites.

Recently, dense connections have attracted substantial attention in computer vision because they facilitate gradient flow and implicit deep supervision during training. Particularly, DenseNet that connects each layer to every other layer in a feed-forward fashion and has shown impressive performances in natural image classification tasks. We propose HyperDenseNet, a 3-D fully convolutional neural network that extends the definition of dense connectivity to multi-modal segmentation problems. Each imaging modality has a path, and dense connections occur not only between the pairs of layers within the same path but also between those across different paths. This contrasts with the existing multi-modal CNN approaches, in which modeling several modalities relies entirely on a single joint layer (or level of abstraction) for fusion, typically either at the input or at the output of the network. Therefore, the proposed network has total freedom to learn more complex combinations between the modalities, within and in-between all the levels of abstraction, which increases significantly the learning representation. We report extensive evaluations over two different and highly competitive multi-modal brain tissue segmentation challenges, iSEG 2017 and MRBrainS 2013, with the former focusing on six month infant data and the latter on adult images. HyperDenseNet yielded significant improvements over many state-of-the-art segmentation networks, ranking at the top on both benchmarks. We further provide a comprehensive experimental analysis of features re-use, which confirms the importance of hyper-dense connections in multi-modal representation learning. Our code is publicly available.

This paper addresses the need for a globally regionalized method for life cycle impact assessment (LCIA), integrating multiple state-of-the-art developments as well as damages on water and carbon areas of concern within a consistent LCIA framework. This method, named IMPACT World+, is the update of the IMPACT 2002+, LUCAS, and EDIP methods. This paper first presents the IMPACT World+ novelties and results and then analyzes the spatial variability for each regionalized impact category. With IMPACT World+, we propose a midpoint-damage framework with four distinct complementary viewpoints to present an LCIA profile: (1) midpoint impacts, (2) damage impacts, (3) damages on human health, ecosystem quality, and resources & ecosystem service areas of protection, and (4) damages on water and carbon areas of concerns. Most of the regional impact categories have been spatially resolved and all the long-term impact categories have been subdivided between shorter-term damages (over the 100 years after the emission) and long-term damages. The IMPACT World+ method integrates developments in the following categories, all structured according to fate (or competition/scarcity), exposure, exposure response, and severity: (a) Complementary to the global warming potential (GWP100), the IPCC Global Temperature Potentials (GTP100) are used as a proxy for climate change long-term impacts at midpoint. At damage level, shorter-term damages (over the first 100 years after emission) are also differentiated from long-term damages. (b) Marine acidification impact is based on the same fate model as climate change, combined with the H+ concentration affecting 50% of the exposed species. (c) For mineral resources depletion impact, the material competition scarcity index is applied as a midpoint indicator. (d) Terrestrial and freshwater acidification impact assessment combines, at a resolution of 2° × 2.5° (latitude × longitude), global atmospheric source-deposition relationships with soil and water ecosystems’ sensitivity. (e) Freshwater eutrophication impact is spatially assessed at a resolution grid of 0.5° × 0.5°, based on a global hydrological dataset. (f) Ecotoxicity and human toxicity impact are based on the parameterized version of USEtox for continents. We consider indoor emissions and differentiate the impacts of metals and persistent organic pollutants for the first 100 years from longer-term impacts. (g) Impacts on human health related to particulate matter formation are modeled using the USEtox regional archetypes to calculate intake fractions and epidemiologically derived exposure response factors. (h) Water consumption impacts are modeled using the consensus-based scarcity indicator AWARE as a proxy midpoint, whereas damages account for competition and adaptation capacity. (i) Impacts on ecosystem quality from land transformation and occupation are empirically characterized at the biome level. We analyze the magnitude of global potential damages for each impact indicator, based on an estimation of the total annual anthropogenic emissions and extractions at the global scale (i.e., “doing the LCA of the world”). Similarly with ReCiPe and IMPACT 2002+, IMPACT World+ finds that (a) climate change and impacts of particulate matter formation have a dominant contribution to global human health impacts whereas ionizing radiation, ozone layer depletion, and photochemical oxidant formation have a low contribution and (b) climate change and land use have a dominant contribution to global ecosystem quality impact. (c) New impact indicators introduced in IMPACT World+ and not considered in ReCiPe or IMPACT 2002+, in particular water consumption impacts on human health and the long-term impacts of marine acidification on ecosystem quality, are significant contributors to the overall global potential damage. According to the areas of concern version of IMPACT World+ applied to the total annual world emissions and extractions, damages on the water area of concern, carbon area of concern, and the remaining damages (not considered in those two areas of concern) are of the same order of magnitude, highlighting the need to consider all the impact categories. The spatial variability of human health impacts related to exposure to toxic substances and particulate matter is well reflected by using outdoor rural, outdoor urban, and indoor environment archetypes. For “human toxicity cancer” impact of substances emitted to continental air, the variability between continents is of two orders of magnitude, which is substantially lower than the 13 orders of magnitude total variability across substances. For impacts of water consumption on human health, the spatial variability across extraction locations is substantially higher than the variations between different water qualities. For regionalized impact categories affecting ecosystem quality (acidification, eutrophication, and land use), the characterization factors of half of the regions (25th to 75th percentiles) are within one to two orders of magnitude and the 95th percentile within three to four orders of magnitude, which is higher than the variability between substances, highlighting the relevance of regionalizing. IMPACT World+ provides characterization factors within a consistent impact assessment framework for all regionalized impacts at four complementary resolutions: global default, continental, country, and native (i.e., original and non-aggregated) resolutions. IMPACT World+ enables the practitioner to parsimoniously account for spatial variability and to identify the elementary flows to be regionalized in priority to increase the discriminating power of LCA.

This paper presents a comparative analysis of different energy management schemes for a fuel-cell-based emergency power system of a more-electric aircraft. The fuel-cell hybrid system considered in this paper consists of fuel cells, lithium-ion batteries, and supercapacitors, along with associated dc/dc and dc/ac converters. The energy management schemes addressed are state of the art and are most commonly used energy management techniques in fuel-cell vehicle applications, and they include the following: the state machine control strategy, the rule-based fuzzy logic strategy, the classical proportional–integral control strategy, the frequency decoupling/fuzzy logic control strategy, and the equivalent consumption minimization strategy. The main criteria for performance comparison are the hydrogen consumption, the state of charges of the batteries/supercapacitors, and the overall system efficiency. Moreover, the stresses on each energy source, which impact their life cycle, are measured using a new approach based on the wavelet transform of their instantaneous power. A simulation model and an experimental test bench are developed to validate all analysis and performances.

In this paper, sliding-mode control is applied on multi-input/multi-output (MIMO) nonlinear systems. A novel approach is proposed, which allows chattering reduction on control input while keeping high tracking performance of the controller in steady-state regime. This approach consists of designing a nonlinear reaching law by using an exponential function that dynamically adapts to the variations of the controlled system. Experimental study was focused on a MIMO modular robot arm. Experimental results are presented to show the effectiveness of the proposed approach, regarding particularly the chattering reduction on control input in steady-state regime.

This work compares the performance of six bias correction methods for hydrological modeling over 10 North American river basins. Four regional climate model (RCM) simulations driven by reanalysis data taken from the North American Regional Climate Change Assessment Program intercomparison project are used to evaluate the sensitivity of bias correction methods to climate models. The hydrological impacts of bias correction methods are assessed through the comparison of streamflows simulated by a lumped empirical hydrology model (HSAMI) using raw RCM‐simulated and bias‐corrected precipitation time series. The results show that RCMs are biased in the simulation of precipitation, which results in biased simulated streamflows. All six bias correction methods are capable of improving the RCM‐simulated precipitation in the representation of watershed streamflows to a certain degree. However, the performance of hydrological modeling depends on the choice of a bias correction method and the location of a watershed. Moreover, distribution‐based methods are consistently better than mean‐based methods. A low coherence between the temporal sequences of observed and RCM‐simulated (driven by reanalysis data) precipitation was observed over 5 of the 10 watersheds studied. All bias corrections methods fail over these basins due to their inability to specifically correct the temporal structure of daily precipitation occurrence, which is critical for hydrology modeling. In this study, this failure occurred on basins that were distant from the RCM model boundaries and where topography exerted little control over precipitation. These results indicate that bias correction performance is location dependent and that a careful validation should always be performed, especially on studies over new regions.

Artificial intelligence (AI) with deep learning models has been widely applied in numerous domains, including medical imaging and healthcare tasks. In the medical field, any judgment or decision is fraught with risk. A doctor will carefully judge whether a patient is sick before forming a reasonable explanation based on the patient's symptoms and/or an examination. Therefore, to be a viable and accepted tool, AI needs to mimic human judgment and interpretation skills. Specifically, explainable AI (XAI) aims to explain the information behind the black-box model of deep learning that reveals how the decisions are made. This paper provides a survey of the most recent XAI techniques used in healthcare and related medical imaging applications. We summarize and categorize the XAI types, and highlight the algorithms used to increase interpretability in medical imaging topics. In addition, we focus on the challenging XAI problems in medical applications and provide guidelines to develop better interpretations of deep learning models using XAI concepts in medical image and text analysis. Furthermore, this survey provides future directions to guide developers and researchers for future prospective investigations on clinical topics, particularly on applications with medical imaging.