Clustering is an unsupervised machine learning method grouping data samples into clusters of similar objects, used as a system support tool in numerous applications such as banking customers profiling, document retrieval, image segmentation, and e-commerce recommendation engines. The effectiveness of several clustering techniques is sensible to the initialization parameters, and different solutions have been proposed in the literature to overcome this limitation. They require high computational memory consumption when dealing with big data. In this paper, we propose the application of a recent object detection Deep Learning model (YOLO-v5) for assisting the initialization of classical techniques and improving their effectiveness on two-variate datasets, leveraging the accuracy and reducing dramatically the memory and time consumption of classical clustering methods.

The ever-growing use of digital systems has led to the accumulation of vast datasets, particularly time series, depicting the temporal evolution of variables and systems. Analysing these time series presents a tremendous challenge due to their inherent complexity and heterogeneity. Addressing an industrial need in the pharmaceutical wholesale sector, this paper introduces a new clustering method for time series: TSCAPE. The TSCAPE method uses a distance matrix calculated using dynamic time warping, followed by multidimensional scaling to project time series into a 2D Euclidean space, thus improving the last clustering stage by K-Means. Unlike conventional techniques, this approach, based on clustering of representation of distances in a Euclidean plane rather than on the curve shape, directly enhances the efficiency of the clustering process. The methodology exhibits significant potential for diverse applications, accommodating varied data types and irregular time series shapes. The research compares multiple variants and proposes metrics to assess their effectiveness on two open-access datasets. The results demonstrate the method's superiority over “only distance comparison clustering techniques”, like dynamic time warping and K-Means, with future prospects aimed at predictive applications and refining the clustering process by exploring alternative, more powerful clustering algorithms.

In the past few years, the increasing number of aircraft in low-altitude traffic has brought new challenges for air traffic controllers. As the ADS-B protocol takes an important part in the air surveillance task, it is important to be prepared to face attacks related to it. Moreover, as the ADS-B is open and uses broadcasts, it is easy for an attacker to emit fake ADS-B data to flood the air traffic controller's screens. In this work, we will present two algorithms, the first one is based on hashing techniques, invariant to geometric transformations to detect trajectory replays. The second one uses a deep-learning model to detect irrelevant ghost trajectories. Our proposed approach can automatically filter a majority of ghost aircraft without risking to filter real aircraft.

Every day the innovative IoT technology is expanding further and further in our environment, with applications deployed in various contexts including cities. Communities can indeed address problems linked to urbanization thanks to this technology through the Smart City concept and thus support a sustainable development of their cities. Artificial intelligence and namely its machine learning branch is expected to reinforce this trend by making smart cities even smarter. However, smart cities can only be successful if they can be trusted and, with this in mind, machine learning can potentially be an efficient tool to mitigate cyberattacks. This paper can be divided into two main parts. In the first part the ability of Gradient Boosting and Deep Learning to make long-term predictions of noise level is studied based on noise data collected in the suburb of an English city. In the second part, we proposed an approach for detecting noise levels anomalies based on predictions. Two types of injections were taken into consideration namely punctual noise level attacks and gradual noise level attacks. Specifically, for the punctual attacks, when the difference between the actual sensed and predicted noise levels is greater than a given threshold, we considered that there is an anomaly in the data. For the gradual attacks, we used a criterion comparing the mean absolute error of predictions in the attacked set of data to statistics of the absolute error in the training set. The obtained results show that our approach, which uses a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) hybrid network for the noise level prediction, can effectively be used to detect the aforementioned types of anomalies. In the case of punctual attacks an increase in sound intensity of 5 dB was detected, while for gradual attacks, smaller changes can be detected.

Nowadays, the number of Internet of things (IoT) devices has rapidly increased due to their increasing use in different real-world applications. The sensor devices represent the basic element of the IoT network because they gather data from various environments and situations, while the sink node serves as the network’s brain because it processes the data and makes decisions. However, the large amount of data that the sensor devices gather and send to the gateway toward the sink, on the one hand, causes the sensor’s limited energy to be depleted and, on the other hand, makes it more difficult to achieve the decisions using these data at the sink. Therefore, before sending data to the gateway, it is important to get rid of any duplicate data while maintaining a high level of data quality. In this paper, a distributed prediction–compression-based mechanism (DiPCoM) for saving power in IoT networks is suggested. DiPCoM makes periodic decisions on sending the data to the gateway. It uses the autoregressive integrated moving average prediction method in each period to predict the next period’s data and decide whether the current data should be sent to the gateway. When the decision is made to send the data to the gateway, an effective compression approach is used by DiPCoM to get rid of the duplicate data. It combines different data transmission reduction techniques such as adaptive piecewise constant approximation, differential encoding, symbolic aggregate approximation, and Lempel–Ziv–Welch. Simulation results based on real-world data show that the DiPCoM method is better than other techniques in terms of energy consumption, data reduction ratio, transferred data size, and data accuracy.

The rise in chronic diseases and the aging of the population led to an increase in the demand for remote healthcare systems that employ biosensors to monitor people's health status. The increasing need for these automated systems has led to the emergence of the Internet of Medical Things (IoMT) networks. In the IoMT networks, the biosensor devices collect vital signs and transmit them to the gateway for further analysis and fusion. In light of the limited biosensor device resources (power, storage, and computation) and the periodical transmission of a large amount of data, it is necessary to optimize the transmission of data in order to conserve power while maintaining data quality at the gateway. Also, it became important to have a decision-making-based machine learning model at the gateway to evaluate a patient's health and make a quick, accurate decision in case of an emergency. This article proposes multibiosensor data sampling and transmission reduction with decision-making (MuDaSaTReD) for remote patient monitoring in the IoMT networks. The MuDaSaTReD achieves this goal on two levels: biosensors and a fog gateway. It uses an energy-saving lightweight data transmission (ELiDaT) algorithm to get rid of the repeated data and then adapts the sampling rate of each biosensor using an adaptive data sampling (ADaS) algorithm. The Multibiosensor Data Sampling and Transmission Reduction with Decision-making (MuDaFuDeC) implements the machine learning model at the fog gateway to learn and decide the situation of the patient according to the received data from the biosensors. The performance evaluation shows that the MuDaFuDeC outperforms other approaches in terms of the data reduction percentage and energy consumption. It keeps a good representation of all the scores at the fog gateway and makes automated, fast, and accurate decisions based on the patient's condition.

BACKGROUND:In acute cardiovascular disease management, the delay between the admission in a hospital emergency department and the assessment of the disease from a Delayed Enhancement cardiac MRI (DE-MRI) scan is one of the barriers for an immediate management of patients with suspected myocardial infarction or myocarditis.OBJECTIVES:This work targets patients who arrive at the hospital with chest pain and are suspected of having a myocardial infarction or a myocarditis. The main objective is to classify these patients based solely on clinical data in order to provide an early accurate diagnosis.METHODS:Machine learning (ML) and ensemble approaches have been used to construct a framework to automatically classify the patients according to their clinical conditions. 10-fold cross-validation is used during the model's training to avoid overfitting. Approaches such as Stratified, Over-sampling, Under-sampling, NearMiss, and SMOTE were tested in order to address the imbalance of the data (i.e. proportion of cases per pathology). The ground truth is provided by a DE-MRI exam (normal exam, myocarditis or myocardial infarction).RESULTS:The stacked generalization technique with Over-sampling seems to be the best one providing more than 97% of accuracy corresponding to 11 wrong classifications among 537 cases. Generally speaking, ensemble classifiers such as Stacking provided the best prediction. The five most important features are troponin, age, tobacco, sex and FEVG calculated from echocardiography.CONCLUSION:Our study provides a reliable approach to classify the patients in emergency department between myocarditis, myocardial infarction or other patient condition from only clinical information, considering DE-MRI as ground-truth. Among the different machine learning and ensemble techniques tested, the stacked generalization technique is the best one providing an accuracy of 97.4%. This automatic classification could provide a quick answer before imaging exam such as cardiovascular MRI depending on the patient's condition.

In the past few years, the fast increase in air traffic load has brought new challenges for air traffic controllers. The air surveillance task has become harder and as a consequence, the actual monitoring tools need to be improved. In this work, a method based on deep learning that automatically detects ADS-B spoofing attacks is proposed. As autonomous drone technologies will, in the near future, be more and more developed, this study focuses on low-altitude traffic. Our tool is based on a classifier model that raises anomalies between true aircraft trajectory shapes and supposed aircraft categories (e.g. planes, helicopters). The proposed approach can detect spoofing attacks with a success rate of 96.2%.

Image obfuscation techniques (e.g., pixelation, blurring and masking,...) have been developed to protect sensitive information in images (e.g. individuals’ faces). In a previous work, we designed a recommendation framework that evaluates the robustness of image obfuscation techniques and recommends the most resilient obfuscation against Deep-Learning assisted attacks. In this paper, we extend the framework due to two main reasons. First, to the best of our knowledge there is not a standardized evaluation methodology nor a defined model for adversaries when evaluating the robustness of image obfuscation and more specifically face obfuscation techniques. Therefore, we adapt a three-components adversary model (goal, knowledge and capabilities) to our application domain (i.e., facial features obfuscations) and embed it in our framework. Second, considering several attacking scenarios is vital when evaluating the robustness of image obfuscation techniques. Hence, we define three threat levels and explore new aspects of an adversary and its capabilities by extending the background knowledge to include the obfuscation technique along with its hyper-parameters and the identities of the target individuals. We conduct three sets of experiments on a publicly available celebrity faces dataset. Throughout the first experiment, we implement and evaluate the recommendation framework by considering four adversaries attacking obfuscation techniques (e.g. pixelating, Gaussian/motion blur and masking) via restoration-based attacks. Throughout the second and third experiments, we demonstrate how the adversary’s attacking capabilities (recognition-based and Restoration & Recognition-based attacks) scale with its background knowledge and how it increases the potential risk of breaching the identities of blurred faces.

Data confidentiality is one of the most critical security services. Many encryption algorithms are currently used to provide data confidentiality. That is why there are continuous research efforts on the design and implementation of efficient cipher schemes. For this purpose, different lightweight cipher algorithms have been presented and implemented on GPUs with different optimizations to reach high performance. Some examples of these ciphers are Speck, Simon which both require less latency compared to Advanced Encryption Standard (AES). However, these solutions require a higher number of rounds but with a more simple round function compared to AES. Therefore, in this paper, a new cipher scheme called “ORSCA” is defined which only requires one round with the dynamic key-dependent approach. The proposed cipher is designed according to the GPU characteristics. The proposed one-round stream cipher solution is suitable for the high data rate applications. According to the performance results, it can achieve high data throughput compared to existing ones, with throughput greater than 5 Terabits/s on a Tesla A100 GPU. Thus, this approach can be considered as a promising candidate for real-time applications. Finally, the security level is ensured by using the dynamic cryptographic primitives that can be changed for each new input message (or for a set of messages: sub-session key). Thus, the proposed solution is a promising candidate for high secure GPU cryptographic algorithms.

In the past few years, smart devices have been rapidly increased due to their ever-increasing use in different real-world applications. Most of these devices are sensor nodes that represents the basic element in the Internet of Things (IoT). This increasing number in sensor devices will lead to an increase in the size of transmitted sensed readings across the internet, spending energy of sensor nodes, and decreasing the lifetime of the network. Therefore, to tackle this problem, an Energy-efficient Transmission Optimization Protocol (ETOP) is proposed to optimize the transmission and the lifetime of Sensor Networks of IoT. ETOP achieves this mission by using a simple reduction algorithm-based correlation clustering at the sensor stage to remove the redundant data before transmitting it to the gateway or sink. The results are conducted using the OMNeT++ simulator which show that the ETOP protocol can optimize the transmission and the lifetime of Sensor networks better than other methods.

Recently, the pervasiveness of multimedia contents raised serious security and privacy concerns, especially with limited devices such as the ones used in the Multimedia Internet of Things (MIoT), which are constrained in terms of computations, memory capacity and power consumption. In this paper, we propose a lightweight format-compliant compression–selection cipher scheme, which could be adopted with different types of image compression algorithms (lossless or lossy). The two proposed cipher variants leverage the compressed data characteristics such as randomness and uniformity, as compared to uncompressed data; this distinguishes the proposed solution from the existing ones. Also, the proposed cipher supports configurable encryption parameters, such as the encryption data rate, to satisfy the requirements of the different MIoT applications, and to optimize the trade-off between security and efficiency. The two proposed variants, which do not require additional operations to produce a format-compliant code-stream, are based on the dynamic key approach, and they require a single round of simple operations. The first variant consists only of one cryptographic operation, permutation or substitution, while the second variant consists of two operations, permutation and substitution. The low number of rounds, combined with simple operations, result into low computational complexity and consequently, low energy consumption and latency. The process for updating the permutation and substitution tables is also lightweight, and it could be implemented in parallel for each sub-compressed part (tile or frequency level), which increases the security level at a minimal computational cost. The proposed substitution, permutation, and substitution–permutations variants exhibit a high throughput with an enhancement of at least 311%, 176%, and 125% compared to the optimized Advanced Encryption Standard (AES) implementation, respectively.

Due to the volume of multimedia sensed data, a network of Multimedia Internet of Things (MIoT) devices faces various challenging constraints, most notably in terms of communication overhead, power consumption, and memory usage. A set of these MIoT devices is unable to overcome the large data-size challenge via the use of the Lossy Multimedia Compression (LMC) such as JPEG and BPG since they are limited in memory and computation. Instead, in this paper, we propose to down-scale images at MIoT devices with a factor of 2, 3 or ≥ 4, which reduces the memory consumption, computation, and communicated data size and consequently the latency and energy consumption. To recuperate high-quality images, we apply a Deep Learning (DL) denoising/super-resolution model at the server-side. On the other hand, as MIoT devices use a wireless connection, there is a higher risk of transmission packets loss compared to a wired connection. Almost, packets loss are managed through costly data re-transmissions or data redundancy. However, these solutions with intrinsically voluminous data such as the multimedia one are costly, especially for limited MIoT devices. To overcome this challenge, the denoising/super-resolution model did also undergo a training model to retrieve high-quality images from down-scaled erroneous ones. The obtained results show how effective this proposed solution is, especially when it comes to the enhancement of visual quality of down-scaled and erroneous images with minimum communication, latency, and consequently resource overhead.

Summary In the modern world, it will be necessary to deploy a large number of sensor devices to sense everything around us in order to detect changes, risks, and hazards and to mitigate them. This increasing number of sensor devices represents an essential data provider in the Internet of Things (IoT). The devices generate and transmit a huge amounts of data which requires a large amount of storage and high processing power to come real‐time processing and speed up the network. It also leads to an increase in high energy consumption. Thus, it is important to remove redundant data to reduce the data transmission before sending it to the gateway while maintaining a good level of data quality. In this paper, a distributed energy‐efficient data reduction (DEDaR) approach based on prediction and compression to minimize the data transmission in IoT Networks is proposed. The DEDaR is used in periods to make decision. In each period, the autoregressive prediction (ARP) is used to predict the data of the next period and make a decision on whether to send the data of the current period to the gateway or not. In the case of data transmission, the redundant data are eliminated using an efficient compression approach based on adaptive piecewise constant approximation (APCA), symbolic aggregate approximation (SAX), and finally fixed code dictionary (FCD) based on Huffman encoding. The simulation results based on real‐sensed data show that the proposed DEDaR approach outperforms the other recent methods in terms of data reduction percentage, transmitted data size, energy consumption, and data accuracy.

Computer simulations of physical phenomena, such as heat transfer, often require the solution of linear equations. These linear equations occur in the form Ax $=\mathbf{b}$ , where A is a matrix, $\mathbf{b}$ is a vector, and $\mathbf{x}$ is the vector of unknowns. Iterative methods are the most adapted to solve large linear systems because they can be easily parallelized. This paper presents a variant of the multisplitting iterative method with convergence acceleration using the Krylov-based minimization method. This paper particularly focuses on improving the convergence speed of the method with an implementation based on the PETSc (Portable Extensible Toolkit for Scientific Computation) library. This was achieved by reducing the need for synchronization - data exchange - during the minimization process and adding a preconditioner before the multisplitting method. All experiments were performed either over one or two sites of the Grid5000 platform and up to 128 cores were used. The results for solving a 2D Laplacian problem of size 1024 2 components, show a speed up of up to 23X and 86X when respectively compared to the algorithm in [8] and to the general multisplitting implementation.

A ceramic monolith risks breakage during production of the exhaust systems in the automotive industry. This is due to its position at a specific angle throughout the canning phase (stuffing technique). To overcome this problem, quality control needs to be automated on each brick. This control aims to adjust, if needed, the positioning of the brick before starting the production. This paper applies image processing techniques following the Canny-Hough method and reaches more than 99% of good detection of straight lines within a tolerance of ±5 degrees, as requested by the plant. Some vision parameters (gain, exposure time and aperture range), have been tested in order to have a better visibility of the reference part. Furthermore, a repeatability test is validated in this paper, allowing the algorithm to be deployed in the plant. The dataset is accessible on the following link: https://doi.org/10.5281/zenodo.5948822

In this work, we propose three efficient variants of a message authentication encryption (MAE) algorithm, which is based on the dynamic key-dependent concept and dynamic operation mode to reach a high level of security. These variants consist of a single pass and a single round, in addition to the use of common operations for the encryption and authentication processes to reduce the required execution time and resources. Accordingly, the proposed scheme outperforms the existing solutions that are based on the static approach with multiple rounds. Furthermore, to reduce the overhead associated with the regeneration of the dynamic key and the corresponding cryptographic primitives, we propose a simple, yet effective update process. In such a scheme, even when the same plaintext is processed, it will be encrypted and authenticated using different cryptographic primitives (substitution and permutation tables in addition to round keys), which guards against the existing cryptanalysis techniques. The experimental results show that the proposed MAE variants are more efficient than the counter with cipher block chaining message authentication code (CCM), Galois message authentication code (GMAC), offset codebook mode (OCB), and the Chacha20-poly1305. The best performance is achieved with the third MAE variant that presents a high throughput with an enhancement of at least 373% compared to CCM, 90% compared to GCM, 23% compared to OCB, and 22% compared to Chacha20-poly1305.

Digital image watermarking has justified its suitability for copyright protection and copy control of digital images. In the past years, various watermarking schemes were proposed to enhance the fidelity and the robustness of watermarked images against different types of attacks such as additive noise, filtering, and geometric attacks. It is highly important to guarantee a sufficient level of robustness of watermarked images against such type of attacks. Recently, Deep learning and neural networks achieved noticeable development and improvement, especially in image processing, segmentation, and classification. Therefore, in this paper, we studied the effect of a Fully Convolutional Neural Network (FCNN), as a denoising attack, on watermarked images. This deep architecture improves the training process and denoising performance, through which the encoder–decoder remove the noise while preserving the detailed structure of the image. FCNNDA outperforms the other types of attacks because it destroys the watermarks while preserving a good quality of the attacked images. Spread Transform Dither Modulation (STDM) and Spread Spectrum (SS) are used as watermarking schemes to embed the watermarks in the images using several scenarios. This evaluation shows that such type of denoising attack preserves the image quality while breaking the robustness of all evaluated watermarked schemes. It could also be considered a deleterious attack.

Quality control is an essential operation for an automotive company like Faurecia. A vast number of references is produced, and many regions of interest need to be checked. For that, quality control is necessary and should be applied to every reference part. Visual inspection is achieved by the operator who checks each part manually. After several checks per day, the operator gets tired and thus may misqualify a welding seam or a component control. To avoid that, Faurecia is trying to integrate automatic quality control to obtain better overall equipment effectiveness (OEE), especially to avoid performance degradation over the operator’s shift. Researches demonstrate the ability of a neural network to reach high precision in detecting object presence or absence. We have been able to achieve an accuracy of 99% with ResNet-50. Apart from accuracy, the other performance matrices used in this work are reliability and cycle time. Our contribution will help the current state of manufacturing by offering an automatic visual inspection, which will lead to other innovative projects in the automotive industry.