International audience

The efficient management of electric vehicle (EV) charging infrastructure is critical to meeting the growing demand for sustainable transportation. This study addresses the Electric Vehicle Charging Scheduling Problem (EVCSP), focusing on maximizing the number of satisfied charging demands. A bi-level optimization framework is developed, with the upper-level problem solved using two approaches: a classical Genetic Algorithm (GA) and a Hybrid Genetic Algorithm (HGA) enhanced with reinforcement learning via Q-learning. The HGA incorporates a Q-table to dynamically guide mutation decisions, balancing exploration and exploitation for improved performance. At the lower level, energy allocation is optimized using a mathematical programming model, ensuring feasibility and compliance with grid and charger constraints. Computational results demonstrate the HGA’s superior ability to handle large-scale instances due to its integration of adaptive learning and optimal energy allocation. The proposed framework advances EV charging management by combining evolutionary algorithms and reinforcement learning to address the complexities of real-world charging scenarios.

The growing adoption of electric vehicles (EVs) demands efficient scheduling of charging resources, particularly in multi-station networks with heterogeneous chargers and limited grid capacities. This study tackles the Multi-Station Electric Vehicle Charging Scheduling Problem (EVCSP), aiming to maximize satisfied charging demands while adhering to operational constraints. A novel mathematical model is proposed, extending single-station frameworks to accurately represent multi-station complexities. To address the scalability challenges of large instances, a hybrid memetic algorithm is developed, integrating evolutionary optimization with mathematical programming for energy allocation at individual stations. Computational experiments reveal that the mathematical model is highly effective for small-scale instances, providing optimal solutions efficiently. For larger scenarios, the hybrid memetic algorithm outperforms the model in scalability and solution quality, demonstrating its robustness for real-world applications. The proposed approaches advance the optimization of EV charging networks, offering scalable and practical solutions for modern transportation systems.

Electric vehicles (EVs) are pivotal to reducing greenhouse gas emissions, yet their growing adoption presents major challenges for power grid management. This paper addresses the NP-hard problem of scheduling EV charging at public stations, considering vehicles’ arrival/departure times and user-defined charging demands. Under power and charger constraints, the goal is to minimize the total deviation between requested and achieved state-of-charge levels. We propose a hybrid optimization framework that integrates a genetic algorithm with mathematical programming. To reduce computational cost, we further introduce a multi-surrogate-assisted approach that approximates the mathematical model. Simulation results demonstrate the effectiveness and efficiency of the proposed methods in solving large-scale EV charging scheduling problems.

Multi-objective optimization problems are characterized by conflicting objectives, making it impossible to identify a single optimal solution. Instead, solution methods aim to produce a diverse set of non-dominated solutions, aka Pareto optimal solutions, each offering different tradeoffs among the objectives. Evolutionary multi-objective algorithms (EMOAs) are commonly employed to generate these varied sets of solutions. However, the abundance of solutions presents a significant challenge for decision-makers (DMs) in identifying the most preferred solution. The problem becomes even more pronounced as the number of objectives increases, requiring exponentially more computational resources and more solutions to properly represent the Pareto optimal set. Interactive EMOAs (iEMOAs) mitigate this challenge by integrating DM preferences into the optimization process to limit the search to regions of the Pareto front that are interesting to the DM. Despite their advantages, existing methods often struggle with effectively learning and utilizing DM preferences. This study investigates tree-based learning methods for preference modeling in iEMOAs by conducting a systematic comparison of decision trees (DTs) and random forests (RFs). Additionally, it examines the impact of active learning as a solution selection strategy for improving preference elicitation. Experimental results demonstrate that RF achieves significantly higher accuracy than DT in learning DM preferences. Furthermore, integrating active learning enhances preference learning within RF, further improving its accuracy. These findings highlight the potential of active learning for enhancing preference-driven optimization, offering more effective strategies for interactive multi-objective decision-making.

The growing adoption of electric vehicles (EVs) necessitates efficient scheduling of charging operations to optimize limited infrastructure and maximize demand satisfaction. This study addresses the Electric Vehicle Charging Scheduling Problem (EVCSP) with the objective of maximizing the number of fulfilled charging requests. We propose two complementary solution frameworks. The first is an enhanced global mathematical programming model that captures the full problem structure and delivers high-quality solutions, even for large-scale instances. The second is a hybrid optimization approach that decomposes the problem into two interrelated components: EV-to-charger assignment and energy delivery optimization. A tabu search (TS) algorithm explores diverse assignment configurations, while each generated assignment is evaluated via exact mathematical programming to optimize energy allocation. Computational experiments show that both frameworks significantly outperform conventional approaches in solution quality and efficiency. These results highlight the potential of the proposed methods for scalable and effective EV charging management, supporting the operational needs of increasingly complex EV networks.

Natural disasters have intensified in recent years, directly affecting populations, especially in urban and semiurban areas where infrastructure is damaged and causes casualties. Rapid damage assessment is essential for decision-making and developing relief measures for victims of such disasters. In the aftermath of such disasters, the application of deep learning to aerial imagery has become increasingly crucial for damage assessment. By categorizing damage levels to buildings and other environmental elements, such as roads, these methods can significantly support informed decision-making processes. RescueNet is proposed as a database for training these intelligent damage estimation models. In this study, we introduce a GA-driven hyperparameter optimization for adapting PSPNet to the RescueNet hurricane imagery dataset. Using a genetic algorithm to explore key design choices (e.g, learning rate, backbone depth, pooling bin sizes, dropout probability, …) we automatically evolve high-performance configurations over successive generations. Our best model boosts mean Intersection over Union on undamaged buildings from 95.16% to 96.57% and on vehicles from 85.97% to 89.76%, outperforming the default PSPNet settings. By sharply improving the model's ability to distinguish intact versus damaged regions, we deliver faster, more reliable disaster-impact assessments. In ongoing work, we are extending this metaheuristic approach to optimize additional critical classes such as roads to further enhance end-to-end operational readiness.

International audience

This study addresses the Electric Vehicle Charging Scheduling Problem (EVCSP) with the objective of maximizing the number of scheduled charging requests while satisfying grid and charger constraints. To solve this problem, a hybrid bi-level optimization framework is developed, where a Graph Convolutional Network (GCN) guides the upper-level assignment of EVs to chargers, and exact methods ensure feasible energy allocation at the lower level. The proposed framework is evaluated on diverse and large-scale instances, with comparisons to a simulated annealing algorithm introduced in the literature as an effective method for solving the studied problem. Results demonstrate that the hybrid GCN-based approach achieves high-quality solutions with improved computational efficiency. These findings underscore the effectiveness of integrating learned graph representations with mathematical programming to solve complex and large-scale EV charging scheduling problems.

Accurate State of Health (SOH) estimation is critical for optimizing battery management, extending lifespan, and preventing unexpected failures in lithium-ion batteries. However, existing methods often face challenges related to computational complexity and data requirements, limiting their practicality in real-time, resource-constrained applications. This paper presents a novel data-driven approach for SOH estimation that combines low computational demands with high accuracy. Using Gaussian Process Regression (GPR) and features extracted from both charge and discharge voltage curves, we achieve precise SOH estimation with a Mean Absolute Error (MAE) of 0.06%. The proposed approach is validated using the Oxford Battery dataset, with results showing consistent accuracy across multiple cells and under varying training data sizes. Notably, the model achieves robust performance even with limited training data, completing estimations in just 0.05 seconds, making it highly suitable for real-time applications. This work contributes to the advancement of battery management systems by providing a computationally efficient, accurate, and real-time SOH estimation method.

In this paper, we present an improved methodology for scheduling electric vehicle (EV) charging at a single charging station, taking into account vehicle arrival and departure times, as well as drivers' charging requirements. The objective is to minimize the difference between desired and final state-of-charge levels while adhering to constraints on power capacity and charger availability. As electric vehicles increasingly contribute to mitigating greenhouse gas emissions, efficient charging strategies become crucial to manage their impact on the electrical grid. Our study delves into the complex task of scheduling EV charging at public stations, where drivers pre-communicate their charging needs. This study proposes improved solution methods, including a standalone mathematical programming model for scheduling, a hybrid heuristic algorithm for assignment, which combines mathematical modeling techniques for evaluation and energy allocation, and a hybrid tabu search algorithm, which uses the same mathematical model employed in the heuristic. Our results demonstrate the effectiveness of our approach in tackling the challenges of EV charging scheduling, highlighting its relevance for sustainable energy management.

This study delves into the electric vehicle charging scheduling problem within a public charging service station. The scheduling task entails managing charging demands from drivers, including arrival and departure times, current battery state-of-charge, and desired state-of-charge at departure. The scheduler must determine whether to accept or reject charging requests based on charger availability and the maximum grid capacity of the station. The primary objective is to minimize the cumulative discrepancy between the desired and final state-of-charge levels for all electric vehicles. To address this challenge, we introduce a novel architecture that harnesses the capabilities of population-based meta-heuristics as a promising method for identifying near-optimal solutions.

This paper introduces a new Distributed State Variable Estimation Algorithm (DSVEA). This algorithm is developed based on the Kalman filter for estimating the state variables of large-scale systems. This new approach involves decomposing the large-scale state estimation problem into several interconnected local sub-estimation problems. To demonstrate its effectiveness, the DSVEA is applied to estimate the state variables of a large-scale power system consisting of five synchronous generators. Simulation results confirm the excellent performance of the DSVEA, validating its potential for practical implementation.

This paper addresses the computationally challenging edge-based facility location problem with the objective of minimizing total travel time while accommodating uniformly distributed demand on network edges. To enhance computational efficiency, the proposed method integrates differential evolution (DE) with three distinct surrogate models: random forest, extreme learning machines, and extreme gradient boosting. While the concept of distributed demand on network edges presents a more realistic depiction of location problems, the necessity of decomposing edges and assigning them to their nearest facilities increases the complexity of the problem at hand. Therefore, the development of an effective and efficient solution method is crucial, particularly in time-sensitive contexts where rapid decisions are essential. Empirical evaluations demonstrate the efficacy and efficiency of the proposed multi-surrogate approach when compared to traditional DE and a leading surrogate-based algorithm. The results illustrate superior computational performance while preserving solution quality across various benchmark functions.

Wireless Sensor Networks (WSNs) represent a key component in smart building systems. An efficient WSN deployment involves selecting the most appropriate positions within the building to place sensors in order to maximize coverage and minimize the deployment cost. This paper proposes a novel approach called the Hybrid Binary Grey Wolf Optimiser (HBGWO) to automate the WSN deployment in indoor environments. The proposed approach integrates the Building Information Modeling (BIM) database to accurately model the physical layout and structural characteristics of the deployment area. Furthermore, a Steiner Tree-based heuristic has been developed to reduce the number of active sensors while preserving the network coverage. Experimental results demonstrate the efficiency and superiority of the HBGWO approach compared to existing methods in literature in terms of network coverage and deployment cost under the connectivity constraint.

Finance, supply chain, healthcare, and energy have an increasing demand for secure transactions and data exchange. Permissioned blockchains fulfilled this need thanks to the consensus protocol that ensures that participants agree on a common value. One of the most widely used protocols in private blockchains is the Practical Byzantine Fault Tolerance (PBFT) which tolerates up to one-third Byzantine nodes, performs within partially synchronous systems and has a superior throughput compared to other protocols. It has, however, an important bandwidth consumption: 2N(N-1) messages are exchanged in a system composed of N nodes to validate only one block. It is possible to reduce the number of consensus participants by restricting the validation process to nodes that have demonstrated high levels of security, rapidity, and availability. In this paper, we propose the first database that traces the behavior of nodes within a system that performs PBFT consensus. It reflects their level of security, rapidity and availability throughout the consensus. We first investigate different Single-Task Learning techniques to classify the nodes within our dataset. Then, using Multi-Task learning techniques, the results are way more interesting with classification accuracies over 98%. Integrating nodes classification as a preliminary step to the PBFT protocol optimizes the consensus. In the best cases, it is able to reduce the latency by up to 94% and the communication traffic by up to 99%.

This study scrutinizes a facility location problem with uniformly distributed demands along the network edges. The objective is to determine the best locations for establishing facilities such that the aggregate traveling time is minimized. Each network edge is divided into two segments, each assigned to its closest open facility. Finding the best combination for establishing facilities and using them as a basis for decomposing network edges form the main decision variables. Due to the NP-hardness of this problem, a Genetic Algorithm is used as the optimization method. This algorithm is known as one of the best metaheuristics for solving this problem. To accelerate the optimization process considering the computationally expensive fitness evaluation of the edge-based location problems, an extreme learning machine is hybridized with the implemented genetic algorithm to serve as a surrogate model for approximating the fitness of the majority of individuals. The results obtained from solving generated instances indicate that while keeping the same quality of solutions, the developed surrogate model-based genetic algorithm significantly reduces the required computational time.

The wireless sensor deployment problem is one of the major issues in wireless sensor networks (WSNs). It involves designing the optimal network topology within the deployment area in order to maximize network coverage and lifetime and minimize cost and energy consumption under the connectivity constraint. The WSN deployment problem is a challenging NP-hard combinatorial optimization problem due to a number of factors, including the size and the type of the deployment area, the number of obstacles, and the number of objectives to optimize. Consequently, metaheuristics are assumed to be the most efficient methods to compute the deployment scheme in a reasonable amount of time. In this paper, several well-known metaheuristics have been tested on the problem of WSN deployment in indoor environments. The problem has been formulated as a constrained single objective optimization problem, and the performance of the selected algorithms has been evaluated through experimentation on a set of ten representative indoor architectural scenarios with varying dimensions and obstacles.

Developing healthy diets in early childhood may help determine future healthy foods. Many kids pass time in childcare, but few studies evaluated the nutritional quality of menus and snacks in childcare homes. Therefore, serving healthier meals is the main phase to attaining that objective. In spite of this, planning a healthful and balanced menu manually is complicated, wasteful, and time-consuming. The objective of this paper is to develop a Multi-period Goal Programming (MGP) model to plan menus for the Tunisian school canteen. School children aged 06 to 12 years old will have three different types of menus according to their cases (standard menus, sports menus, and diabetics menus). This model minimizes the sum of deviations of nutrient requirements for three groups of menus and for one week (Monday, Tuesday, Wednesday, Thursday, and Friday). The MGP is broken down into a number of periods, and an ideal choice needs to be made on the relationship between these decisions in each period. This model respects the budget provided by the canteen manager, it ensures assigning the recipes according to their correct category (sweet snack, salad, main meal, fruit and salty snack), it guarantees that each recipe for each menu is not used more than once per week as well as nutrients requirement constraint. As a consequence, the final dietary plan for the children is an optimal assortment of the recipes intake, related to minimizing deviations from the explained goals. The diet model develops a nutritional adviser's work and avoids errors when preparing a diet plan manually.

Wireless sensor network connectivity is a crucial parameter since it keeps the network operative. Network connectivity may be lost due to a variety of factors, such as energy depletion and sensor node failure. Therefore, the network will be partitioned into a set of disjoint sets, resulting in a loss of data collected by isolated sets. In this paper, we address the problem of connectivity repair for stationary sensor networks (WSNs) in case of multiple disjoint partitions. We propose two heuristics based on Dijkstra algorithm and minimum Steiner tree respectively, to deploy the minimum number of additional nodes while preserving the initial topology. For the two heuristics, a procedure is executed in the first stage to merge disjoint sets having a shared zone in their neighboring deployment zones to reduce the complexity of the solution. The first heuristic is adapted to free-obstacle areas and areas with few obstacles. It connects the less distant segments using Dijkstra algorithm. The second heuristic is rather appropriate for areas with opaque obstacles. Simulation experiments validate the effectiveness of the proposed methods compared to existing approaches.

This paper presents a novel bi-level optimization approach for addressing the 3D truck loading problem, incorporating considerations of axle weights and items with diverse delivery time windows, as defined in the 2022 EURO/ROADEF Challenge. The main goal is to optimize the efficient stacking and allocation of items into appropriate trucks, to minimize both inventory costs and transportation expenses, which encompass the additional costs associated with employing extra trucks. The first level focuses on optimizing the assignment of items to appropriate trucks, considering factors such as weight limits and compatibility. The second level initiates using an efficient heuristic for generating stacks. Using a combination of a developed Tabu Search algorithm with multiple heuristics, the generated stacks are placed within the assigned trucks such that the unused space is minimized. The computational results demonstrate the advantage of our proposal results compared to the best-known results from the EURO/ROADEF challenge, considering identical system specifications and computational times.

This paper addresses the facility location problem in the context of multiple-server facilities subject to congestion. The objective is to select a subset of facilities from a pool of candidate locations in order to meet customers’ demands. Additionally, the number of servers allocated to each facility is treated as a decision variable, and the service time for each server follows an exponential distribution. As network location problems are known to be NP-hard, this study introduces a random forest as a surrogate model with differential evaluation to minimize the aggregate expected traveling times and aggregate expected waiting times of customers. The proposed algorithm is implemented and evaluated on a set of test problems with different sizes and specifications, demonstrating its high efficiency compared to differential evaluation.

Nowadays, health care has become a constant concern of all countries around the world, especially after the emergence of the Coronavirus (COVID-19) and all its variants. Billions of dollars are being paid through the World Health Organization to improve health care. Scientific research laboratories play a pioneering role in this area as well. Due to the importance of information related to the patient and his medical history, it is necessary to exchange these information between various health centers in order to be better treated. Security in Health care information exchange (HIE) plays an important role because different healthcare facilities (HCFs) exchange sensitive data which can affect the patient’s privacy. Researchers propose many approaches in order to enhance security and maintain privacy. They also try to solve many drawbacks in this field like efficiency, accuracy, and scalability. Unfortunately, all the proposed techniques tackle some parameters and drop other ones. To decide which blockchain-based approach is more efficient for HIE systems, by evaluating each approach’s effectiveness based on its security, integrity, privacy, accuracy, scalability, efficiency, and latency qualities, we present a comparative analysis between a number of recent approaches in the HIE sector. Further, we use real patients’ data to measure each parameter, then we apply the Friedman test on the obtained results for each approach.

Nowadays, Healthcare Information Exchange (HIE) plays a vital role in healthcare systems; it allows healthcare providers to access and share patient medical data electronically and securely. Subsequently, HIE eliminates redundant or unnecessary testing, and improves public health reporting and monitoring. Security is a very important challenge in the HIE systems since data are exchanged between different healthcare facilities (HCF), thus, the data are subject to be modified or altered. Hence, detecting modified or risky transactions is becoming a fundamental operation in HIE systems. In this paper, we propose a secure framework that combines between deep learning and blockchain, called as DeepChain, for detecting risky transactions in HIE systems. On one hand, DeepChain uses two types of blockchain to enhance the data security: a blockchain to store ordinary patient data, and an off-chain to store the sensitive patient's data. On the other hand, DeepChain uses an advanced deep learning model called generative adversarial network (GAN) with two-folds: first, it enhances the training phase of the model by generating additional synthetic health data, then it uses a discriminator to accurately detect the risky transactions in the testing phases. We evaluated the performance of our framework based on real health data while the obtained results shows the efficiency of DeepChain in detecting risky transactions and enhancing HIE security.

This study proposes a new bi-objective mixed-integer non-linear mathematical model for an interruptible single-server congested facility location problem with uniformly distributed demands along the network edges. It is assumed that in the case of server disruption, all the waiting customers leave the facility without receiving the service, and there would be no entry until fixing the server. Limiting by the maximum waiting time threshold, this study aims to determine the number and locations of established facilities. The first objective function minimizes the facility establishment costs, while the second objective function is to minimize the aggregate traveling, waiting, and demand lost costs. Due to the NP-hardness nature of the problem, several state-of-the-art evolutionary multi-objective optimization (EMO) algorithms are applied to find the set of non-dominated solutions. The results indicate that the applied SPEA - II algorithm outperforms its competitors in the majority of generated test cases.

As one of the most prominent variants of the facility location problem, the p-median problem aims to determine the best locations for establishing p number of facilities such that the aggregate customers' transportation cost is minimized. Since the p-median problem is classified as NP-hard, the application of metaheuristics to solve it is inevitable. Considering the fast development in metaheuristics, choosing the most appropriate algorithm to solve this problem is a difficult task. Therefore, this work presents a comparative study of several classical and recently developed nature-inspired optimization algorithms to solve the discrete uncapacitated p-median problem on several randomly generated test instances with different sizes and spec-ifications.

As smart cities become more decentralized, the need for reliable and secure cyber-physical systems (CPS) that guarantee safe interactions and secure data storage without loss of privacy is continuously increasing. Blockchain is a rapidly emerging technology in this domain. It demonstrated effectiveness thanks to the cryptographic mechanisms it utilizes and to its immutability. Private blockchains are the most suited to applications that require privacy and confidentiality when data is very sensitive. In this case, the most commonly used consensus protocol is Practical Byzantine Fault Tolerance (PBFT). However, PBFT requires the participation of all nodes in the consensus process, which increases bandwidth consumption and consensus delay significantly. In this paper, we propose an adaptive PBFT protocol called APBFT that optimizes the number of nodes participating in the consensus based on their response time and credibility. Therefore, we reduce the amount of communication and the response delays. We maintain the asynchrony of the algorithm so that it remains resilient to DoS attacks. The simulation results show that our algorithm outperforms the original PBFT in terms of delays and message traffic.

Healthcare requires the cooperation of many administrative units and medical specialties. The Internet of Things (IoT) is involved into healthcare field and plays an extremely important role by providing healthcare services. In the Internet of Healthcare Things (IoHT) several challenges appeared in terms of limited battery life, long processing time, large amounts of collected data, paquets overhead on sink,…etc. A lot of research studies have been done in order to improve the healthcare IoT based applications. However, these proposed systems have been focused on some specific purpose without ensuring an effective solution for all problems. This paper presents a comparative study for the recent advances in Internet of healthcare monitoring. In addition, an optimized intra WBSN communication (OIC) an is proposed in order to overcome the over-mentioned challenges. OIC is based on merging an energy efficient routing protocol with the data transmission process. This approach also reduces the data acquisition redundancy, and finds the optimum path for data transmission, according to the patient situation. Our approach is implemented and tested on real biosensor data while the obtained results prove the efficiency on extending the network lifetime (up to 84%), removing the data transmission redundancy (up to 60%), minimizing the overhead on sink (up to 83%), and optimizing the transmission time. Based on a comparative study with four state-of-the-art methods, the superiority of our approach is validated.

This study proposes a novel multi-objective integer programming model for a collision-free discrete drone path planning problem. Considering the possibility of bypassing obstacles or flying above them, this study aims to minimize the path length, energy consumption, and the accumulated maximum path risk simultaneously. The static environment is represented as 3D grid cells. Due to the NP-hardness nature of the problem, several state-of-the-art evolutionary multi-objective optimization (EMO) algorithms with customized crossover and mutation operators are applied to find a set of non-dominated solutions. The results show the effectiveness of applied algorithms in solving several generated test cases.

The amount of scientific conferences and journal articles continues to increase and new approaches are required to support users in finding relevant publications. This study investigates to what extent a new machine learning (ML) pipeline may preferentially identify links between similar scientific articles. The characteristics of intersections and unions of keywords, contextualized keywords (i.e., synsets) and neighbors are computed and used to train a ML model. Automated machine learning (AutoML) is then applied to ease the search for a new pipeline. Extensive experiments demonstrated that a newly designed ML model achieves an accuracy of 90% on a dataset of approximately 120,000 article pairs. These results suggest that application of ML for proposing new recommendation systems could have in the long term a positive impact in the literature.

The optimal camera placement problem is that of determining the best possible set of camera positions and orientations in order to meet application-specific constraints and objectives. This paper focuses on one application of the problem: global area surveillance. Given an area to be covered, the question is to design a camera network which fully covers critical subareas and proceeds in a best-effort manner in the rest of the environment, given a limited budget. This is achieved through the integration of user-provided input into a mixed combinatorial model which brings together two variants of a popular optimisation problem. Time-efficient algorithms then allow for regular user interaction in between solving iterations. This human-assisted design is based off requirements set by experts and decision makers and yields components of a decision support system to support law enforcement officers and officials when designing video surveillance infrastructure.

This paper considers multiobjective UAV path planning in a real 3D environment with the objective to find a safe energy-efficient path. An Enhanced Non-dominated Sorting Genetic Algorithm-II, called ENSGA-II, is proposed and combines several sorts of heuristic information to customize crossover and mutation operators. Furthermore, a local search and a ranking-based roulette wheel selection are incorporated for the mating procedure. Experiment results confirm that ENSGA-II has a better convergence rate and spread of solutions on several new real-world datasets. The effectiveness of the local search component is also validated on the CrazyS robot operating system (ROS) package which consists of a pelican quadcopter's modeling.

Neural architecture search (NAS) has achieved great success in different computer vision tasks such as object detection and image recognition. Moreover, deep learning models have millions or billions of parameters and applying NAS methods when considering a small amount of data is not trivial. Unlike computer vision tasks, labeling time series data for supervised learning is a laborious and expensive task that often requires expertise. Therefore, this paper proposes a simple-yet-effective fine-tuning method based on repeated k-fold cross-validation in order to train deep residual networks using only a small amount of time series data. The main idea is that each model fitted during cross-validation will transfer its weights to the subsequent folds over the rounds. We conducted extensive experiments on 85 instances from the UCR archive for Time Series Classification (TSC) to investigate the performance of the proposed approach. The experimental results reveal that our proposed model called NAS-T reaches new state-of-the-art TSC accuracy, by designing a single classifier that is able to beat HIVE-COTE: an ensemble of 37 individual classifiers.

In today's time, we live in the big data era where every event and thing about us is monitored and registered for a later analysis. In addition, such amount of big data is collected with a large speed (velocity) and does not fit a fixed structure (unstructured type). One of the most contributors of big data in this era is wireless multimedia sensor network (WMSN). Typically, WMSN consists of a set of sensors that collect three types of data about a zone of interest: numerical, images and videos. Indeed, the big data collected in WMSN along with the density deployment of network, especially in large-scale zones, provide real challenges for the end users in terms of data storage and processing. In this paper, we propose an efficient and robust Hadoop-based framework for big data collection, processing and storage in WMSN. The proposed framework relies on Hadoop ecosystem tools and introduces two fault detection algorithms (moving average and exponential smoothing) in order to preprocess data before storage. Through real sensor data with various types, we show the effectiveness of our framework in terms of processing storage speed and regenerating of missing data.

Differential Evolution algorithm (DE) has been investigated in several studies. Indeed, it has been revealed that despite its successful search operators, DE may get trapped in local optimum due to the poor parameter configuration, and the inappropriate search operators. In this study, we introduce a resilient mutation strategy well-suited to real-world problems. Moreover, a machine learning-based parameter adaptation mechanism is proposed to configure DE parameters during the search process. The new adaptive DE has been tested to find the optimal mechanical structure of a novel electric motor topology. Furthermore, the results have been validated using the real-world problems from the CEC 2011 test suite. The results have revealed that the proposal can be competitive compared to recent adaptive DE algorithms.

Optimal camera placement (OCP) is one of many practical applications of a core NP-complete problem in the field of combinatorial optimisation: set cover (SCP). In a generic form, the OCP problem relates to the positioning and setting up of individual cameras such that the overall network is able to cover a given area while meeting a set of application-specific constraints (such as image quality or redundancy) and optimising an objective, typically minimum cost or maximum coverage, depending on the application's focus. In this paper, we consider the problem of positioning and orienting a minimal number of cameras such that the network is able to reach full coverage. More specifically, we introduce a framework for OCP instance generation which leaves the common realm of academic study cases and models the problem in real-world settings, using 8 West-European cities for numerical tests. A baseline is established by running several basic algorithms, which serve as a starting point for a more focused benchmark involving various state-of-the-art algorithms from both OCP and SCP literature. The results are then discussed and several elements highlighted for future research.

Differential Evolution (DE) is a well-known metaheuristic designed to solve continuous optimization problems. Its simple structure and straight forward search operators make it suitable for solving a wide range of real world problems. Despite its success, DE performance may be limited when tackling high dimensional complex problems. Therefore, its algorithmic structure can be reconsidered by adaptively controlling its parameters, and incorporating more resilient search operators. In this study, a Q-learning-based strategy is proposed to adapt DE parameters during the search process. Moreover, an eigenvector-based crossover is introduced in order to accelerate the convergence rate when ill-conditioned landscapes are treated. However, to avoid premature convergence, a simple yet efficient switching technique is proposed to choose between the normal and the eigenvector-based crossover. Due to the high computational time that might occur when applying the eigenvector-based crossover, a parallel counterpart of the algorithm has been implemented using graphics processing units (GPUs). The proposed algorithm has been applied to find the optimal mechanical structure of a recent electric motor. Its performance has been also validated by testing the proposal on CEC 2011 test suite, which contains a set of real world problems. The experimental results reveal the competetive performance of our algorithm compared to recent adaptive DE versions. Besides, the parallel version of the proposal achieved a serious speedup compared with the sequential version while keeping the same results.

Deep neural networks have revolutionized many fields such as computer vision and natural language processing. Inspired by this recent success, deep learning started to show promising results for Time Series Classification (TSC). However, neural networks are still behind the state-of-the-art TSC algorithms, that are currently composed of ensembles of 37 non deep learning based classifiers. We attribute this gap in performance due to the lack of neural network ensembles for TSC. Therefore in this paper, we show how an ensemble of 60 deep learning models can significantly improve upon the current state-of-the-art performance of neural networks for TSC, when evaluated over the UCR/UEA archive: the largest publicly available benchmark for time series analysis. Finally, we show how our proposed Neural Network Ensemble (NNE) is the first time series classifier to outperform COTE while reaching similar performance to the current state-of-the-art ensemble HIVE-COTE.

Time Series Classification (TSC) problems are encountered in many real life data mining tasks ranging from medicine and security to human activity recognition and food safety. With the recent success of deep neural networks in various domains such as computer vision and natural language processing, researchers started adopting these techniques for solving time series data mining problems. However, to the best of our knowledge, no previous work has considered the vulnerability of deep learning models to adversarial time series examples, which could potentially make them unreliable in situations where the decision taken by the classifier is crucial such as in medicine and security. For computer vision problems, such attacks have been shown to be very easy to perform by altering the image and adding an imperceptible amount of noise to trick the network into wrongly classifying the input image. Following this line of work, we propose to leverage existing adversarial attack mechanisms to add a special noise to the input time series in order to decrease the network's confidence when classifying instances at test time. Our results reveal that current state-of-the-art deep learning time series classifiers are vulnerable to adversarial attacks which can have major consequences in multiple domains such as food safety and quality assurance.

Transfer learning for deep neural networks is the process of first training a base network on a source dataset, and then transferring the learned features (the network’s weights) to a second network to be trained on a target dataset. This idea has been shown to improve deep neural network’s generalization capabilities in many computer vision tasks such as image recognition and object localization. Apart from these applications, deep Convolutional Neural Networks (CNNs) have also recently gained popularity in the Time Series Classification (TSC) community. However, unlike for image recognition problems, transfer learning techniques have not yet been investigated thoroughly for the TSC task. This is surprising as the accuracy of deep learning models for TSC could potentially be improved if the model is fine-tuned from a pre-trained neural network instead of training it from scratch. In this paper, we fill this gap by investigating how to transfer deep CNNs for the TSC task. To evaluate the potential of transfer learning, we performed extensive experiments using the UCR archive which is the largest publicly available TSC benchmark containing 85 datasets. For each dataset in the archive, we pre-trained a model and then fine-tuned it on the other datasets resulting in 7140 different deep neural networks. These experiments revealed that transfer learning can improve or degrade the models predictions depending on the dataset used for transfer. Therefore, in an effort to predict the best source dataset for a given target dataset, we propose a new method relying on Dynamic Time Warping to measure inter-datasets similarities. We describe how our method can guide the transfer to choose the best source dataset leading to an improvement in accuracy on 71 out of 85 datasets.

The performance of Differential Evolution (DE) algorithm strongly depends on its control parameters. Despite its efficiency and wide use, it might get trapped in local minimum due to premature convergence. In this study, a novel parameter adaptation strategy is proposed to address the mentioned problems. To do so, a pheromone matrix is employed to adjust parameter setting of the algorithm during the optimization process. Moreover, the convergence issue of DE is tackled by incorporating a new restart strategy. The performance of the proposed algorithm is firstly evaluated on the CEC 2011 real world problems test suite. Thereafter, we applied the algorithm to find optimized structure of a recent electric motor design considered for this study. The results reveal the competitive performance of the proposed approach with state-of-the-art algorithms.

The surrogate models are offered as effective tools to approximate computationally expensive objective functions. This study investigates how approximation strategy of such models can be used for high dimensional protein structure prediction (PSP) problems. Two major contributions of the proposed approach are: 1) employing Stochastic Response Surface (SRS) to bias the initial population toward promising areas and 2) using queries of an active learning algorithm and a surrogate model to replace in part the original computationally expensive solver. The introduced framework is applied on several extensions of the differential evolution (DE) algorithm which are among noteworthy approaches for the PSP. Numerical experiments indicate that the proposed schema is able to improve performance of the conventional algorithms for the PSP problems in both terms of convergence speed and accuracy.

The majority of evolutionary algorithms (EAs) adopt Pseudo-Random Numbers Generator (PRNG) to initialize their population. This can affect the behavior of an EA for high dimensional problems due to the curse of dimensionality and has been known as a potentially serious challenge. Therefore, intelligent initialization of individual candidates has been more explored recently. As a different approach, this study proposes a machine-learning based algorithm to address the aforementioned problem. The introduced SVM based Smart Sampling, we call as SVM-SS, employs Support Vector Machine (SVM) to discover promising regions faster. The proposed method and Differential Evolution (DE) are then combined to evaluate our approach. Numerical results on a set of classic benchmark functions show that the proposed algorithm performs better in comparison with several state-of-the-art population initialization methods. To examine the scalability of the SVM-SS, it is also applied on large scale optimization problems and such results were also in consonance with the previous experiments.

Zeolite structure determination is an interesting challenge even with the progress in terms of structural resolution from X-rays and electron diffraction. The infinite number of potential solutions and the computational cost of this problem make the use of an evolutionary algorithm significant for this challenge. In this paper, we propose a new parallel and distributed hybrid genetic algorithm called MEmory Genetic Algorithm Hybridized for Zeolite (MEGA-HZ). This experimentation shows that the proposed algorithm is able to satisfy the constraints of the objective function to determine viable zeolite structures. From the 6 unit cell parameters and density, the MEGA-HZ has found 6 different viable zeolite structures.

The works on Search Based Software Engineering (SBSE) have been a big increase in the last decade. An approach to software engineering in which search based optimisation algorithms are applied to address problems in software engineering. SBSE has been applied to problems throughout the software engineering lifecycle, from requirements and project planning to maintenance and re-engineering. This paper provides a modification and an implementation of SBSE on evolutionary multi-objective based approach for deployment of wireless sensor network (WSN) with the presence of fixed obstacle. In this work a multi-objective evolutionary algorithms based on elitist non-dominated sorting genetic algorithm (NSGA-II) is proposed to address the deployment problem. Two functions namely ranking function and fitness function are used to select the best optimal solution from Pareto optimal fronts.

The quadratic assignment problem (QAP) is one of the most studied NP-hard problems. It is a problem known for its computational cost which makes it a good candidate for parallel and distributed design. In this paper, we propose a new Distributed Multistart Hybrid Iterative Tabu Search (DMHITS). This algorithm follows the design of the algorithmic level. Through 34 of the hardest well-known instances from QAPLIB benchmark, the DM-HITS can get the best known solution for almost all the instances. From the 340 runs on these benchmark instances, our algorithm gets more than 300 times the best known solution. This experimentation shows that our proposed algorithm can exceed or equal six leading algorithms from the literature.

In this paper, a study about wireless sensor network (WSN) deployment strategy is demonstrated and made workable for the use of multi-objective approach. The development of sensor nodes by considering multiple objectives and existence of fixed obstacles is an important optimization problem. There are two objectives in this study, connectivity and coverage as two fundamental issues in wireless sensor networks deployment. In this work a multi-objective evolutionary algorithms based on elitist non-dominated sorting genetic algorithm (NSGA-II) is proposed to address this problem. Two proposed functions, ranking function and fitness function, are used to determine the best optimal solution from Pareto optimal fronts. Further we presented simulation and analysis to verify and validate the deployment of wireless sensor network in area with the presence of permanent obstacles.

This paper is a study about deployment strategy for achieving coverage and connectivity as two fundamental issues in wireless sensor networks. To achieve the best deployment, a new approach based on elitist non-dominated sorting genetic algorithm (NSGA-II) is used. There are two objectives in this study, connectivity and coverage. We defined a fitness function to achieve the best nodes deployment. Further we performed simulation to verify and validate the deployment of wireless sensor network as an output from the proposed mechanism. Some performance parameters have been measured to investigate and analyze the proposed sensor-deployment. The simulation results show that the proposed algorithm can maintain the coverage and connectivity in a given sensing area with a relatively small number of sensor nodes.

Imperialist competitive algorithm (ICA) is a population based metaheuristic inspired from imperialistic competition among empires. In order to improve its performances, we propose to hybridize ICA with the Nelder-Mead simplex method. The simplex algorithm is run if a stagnation criterion is satisfied, in order to help ICA escape local optima and to improve its intensification capabilities. The proposed hybrid ICA-simplex algorithm, called ICAS, is first analyzed and compared to the unmodified ICA and two well-known algorithms using the benchmark functions provided during the 2005 IEEE Congress on Evolutionary Computation. The results show the efficiency of the proposed hybrid algorithm. Then, it is used to optimize the design of a permanent-magnet machine used to motorize an electric scooter. The solution found by ICAS is shown to be better than those of several well-known metaheuristics.

When a local optimal solution is reached with classical Particle Swarm Optimization (PSO), all particles in the swarm gather around it, and escaping from this local optima becomes difficult. To avoid premature convergence of PSO, we present in this paper a novel variant of PSO algorithm, called MPSOM, that uses Metropolis equation to update local best solutions (lbest) of each particle and uses mutation operator to escape from local optima. The proposed MPSOM algorithm is validated on seven standard benchmark functions and used to solve the problem of reducing memory energy consumption in embedded systems (Scratch-Pad Memories SPMs). The numerical results show that our approach outperforms several recently published algorithms.

We propose a novel algorithm called GA-MDP for solving the frequency assignment problem. GA-MDP inherits the spirit of genetic algorithms with an adaptation of Markov decision processes (MDPs). More precisely policy iteration (PI) and value iteration (VI) are used as mutation operators. Experimental results show that for our application, GA-MDP that uses PI as a mutation operator improves the quality and time performances of the hybrid algorithms and hybrid MDP designed previously by the authors for solving the same problem

This paper presents a new hybrid genetic algorithm used to solve a frequency assignment problem. The hybrid genetic algorithm presented in this paper uses two original mutation operators. The first mutation operator is based on a greedy algorithm and the second one on an original probabilistic tabu search. The results obtained by our algorithm are better than the best known results obtained by other methods like tabu search and hybrid genetic algorithm. Our results are validated in the field of radiobroadcasting and compared to the best existing solutions in this domain.