Abstract The interplay between machine learning and quantum computing can lead to unprecedented prospects for both fundamental research and real-world applications. In this new field of quantum machine learning, quantum classifiers, i.e., quantum algorithms for solving classification problems, have attracted the most attention. However, despite the promising performance of these new models for classification tasks, they are complex black-box models. Therefore, like or even more so than classical machine learning models, their use in critical contexts such as medical diagnosis could be hampered without the support of explanations. This paper addresses this challenge by investigating the use of Local Interpretable Model-agnostic Explanations (LIME) for quantum classifiers, with a specific focus on the stability of the explanations. Stability is a key property for ensuring that explanations are consistent and trustworthy, especially when decisions impact sensitive domains. The paper conducts a systematic stability study of LIME in the quantum setting, validating whether a classical explainable AI technique can be meaningfully applied to quantum models. As shown in experiments involving different datasets belonging to critical applications, LIME provides explanations for quantum support vector machines and quantum neural networks that are not only interpretable but also stable. By highlighting the stability of explanations, these findings demonstrate the suitability of LIME as an explainability tool for quantum classifiers.

Blood is one of the most important and common types of physical evidence left at crime scenes and used by forensic investigators to determine the dynamics of violent crimes. Therefore, it is important to be able to detect blood stains at the crime scene but without contaminating it. For this reason, recently, classifying hyperspectral images has gained popularity in the procedures to detect blood at the crime scene with respect to traditional approaches based on chemical tests. However, defining a classifier that distinguishes between blood spectra and all the other possible ones requires to train machine learning models with all possible classes of substances left at a crime scene. Since this is no a realistic task, the current approaches perform a classification between blood spectra and a limited number of other substances. In order to overcome this limited vision, this paper proposes, for the first time, to detect blood at the crime scene using one-class classification methods that have the feature of being trained only with samples of the target class (in our case, blood class). In detail, the main goal of this paper is to compare several one-class classification methods and analyse their performance. As shown in the experimental session involving a publicly available hyperspectral-based bloodstain dataset, one-class classification methods result in accurate techniques to detect blood at the crime scene.

In the Noisy Intermediate-Scale Quantum (NISQ) era, efficient quantum circuit synthesis is essential for optimizing limited qubit resources and mitigating noise effects. Quantum Linear Reversible Circuits (QLRCs) are a fundamental class of circuits with applications in quantum compilation or quantum error correction. They are composed exclusively of CX and SWAP gates, which are among the noisiest operations on current quantum hardware. Therefore, their efficient synthesis is pivotal for executing QLRCs on NISQ processors, which have restricted connectivity between qubits. In this direction, we introduce a Reinforcement Learning (RL)-based synthesis approach using Proximal Policy Optimization (PPO) to synthesize QLRCs on generic topologies of up to five qubits. This work extends previous work that did not handle different topologies of quantum processors. Our method dynamically adapts to hardware constraints, learning optimal or near-optimal gate decompositions. We compare our approach against Qiskit’s synthesis methods for QLRCs, demonstrating that our RL-based synthesizer consistently achieves lower CX gate depth. These results highlight the potential of RL-driven quantum circuit synthesis as a powerful alternative to traditional heuristic-based techniques, paving the way for more efficient AI-based quantum compilation strategies in the NISQ era.

Recently, quantum computing has emerged as a new paradigm that promises to improve artificial intelligence techniques. One of the research fields that is certainly benefiting from this new computational paradigm is evolutionary optimization. In literature, efforts have been already made to run evolutionary algorithms on quantum computers using quantum effects such as superposition and entanglement to converge towards sub-optimal solutions of hard problems. However, the performance of these quantum evolutionary approaches is limited by the number of qubits available on current quantum devices. This limitation is more noticeable in the case of continuous optimization problems, where the search space is potentially infinite. This paper presents a recursive algorithmic structure that embodies a quantum evolutionary algorithm to overcome the limitations mentioned above. The result is an innovative and efficient approach in the context of quantum evolutionary optimization.

The nuclear shell model is one of the most adopted many-body methods for the description of atomic nuclei whose main ingredient is the effective Hamiltonian. One of the approaches widely used to derive it is of phenomenological type, where its matrix elements are considered parameters to be fixed to reproduce the experimental data. However, the number of parameters as well as the number of experimental data dramatically increases with the mass of the nuclei under investigation and the commonly adopted procedures such as least-square fitting become computationally prohibitive. Therefore, there is a strong emergence in finding alternative approaches to construct effective Hamiltonians for heavy nuclei. To pave the way in this direction, the proposed work exploits for the very first time Genetic Algorithms. Indeed, their ability to simultaneously examine and manipulate sets of possible solutions could be crucial to deal with the above issues. The suitability of Genetic Algorithms in computing effective Hamiltonians is evaluated experimentally for the p - shell interaction, where the obtained results, without any physical constraint on the parameters, outperform the current widely-adopted description represented by the Cohen and Kurath solution.

The regulation of traffic lights in a signalised urban network requires optimizing objective functions that represent performance indicators of one or more intersections (such as delay or queue length). In this scenario, evolutionary algorithms are adopted to find suitable approximate solutions, in cases when no deterministic algorithm for finding the exact solution is known. This paper attempts to further improve the performance of evolutionary approaches by using a hybrid quantum-classical genetic algorithm to find the optimal configuration of the green signal timing regulating the traffic flow across two interacting junctions. The adopted algorithm, run on IBM quantum computer simulators, is shown to be suitable for the optimization problem at hand. Indeed, the experimental results highlight some of the strengths of the proposed technique with respect to the purely evolutionary approach, and encourage the application of this approach to more complex and close-to-real application scenarios.

Quantum machine learning is a research area that explores the interplay of ideas from quantum computing and machine learning to speed up the time it takes to train or evaluate a machine learning model. Variational quantum classifiers are among the most widely used quantum models for supervised machine learning and base their operation on learning free parameters through conventional optimization algorithms. In spite of their potential benefits, there is a design issue making quantum variational classifiers not fully operative: the existence of exponentially vanishing gradients, known as “barren plateau landscapes”. A barren plateau is a trainability problem that occurs in optimization algorithms for quantum machine learning when the problem-solving space turns flat as the algorithm is run. In that situation, the algorithm cannot find the downward slope in what appears to be a featureless landscape and there's no clear path to the optimum of the cost function. In this challenging scenario, evolutionary optimization techniques can be a viable choice in solving the problem because of their gradient independence. This paper introduces a comparative study involving different evolutionary algorithms to assess their suitability in acting as optimizer for a specific quantum machine learning method, known as variational quantum classifier.

Recently, a significant interest is arising in developing techniques capable of correcting errors in noisy quantum computations without using additional quantum resources. These techniques, known as mitigation methods, are aimed at post-processing the quantum outcome without working at quantum hardware level. Among the most error-prone operations on the current quantum devices, there is surely the quantum measurement. Currently, the most popular mitigation method for quantum measurement error consists of computing a so-called mitigation matrix to be applied on the results outputted by a quantum processor to transform them and make them closer to the ideal ones. In this paper, a new measurement error mitigation method based on a competent memetic algorithm is proposed to generate an appropriate mitigation matrix. As shown in the experimental session, the proposed measurement error mitigation method shows better results when compared with the conventional method and the state of the art among the evolutionary approaches.

Quantum computers can be a revolutionary tool to implement inference engines for fuzzy rule-based systems. In fact, the use of quantum mechanical principles can enable parallel execution of fuzzy rules and allow them to be used efficiently in complex contexts such as distributed and big data environments. This article introduces the very first quantum-based fuzzy inference engine that is capable of providing exponential acceleration in fuzzy rule execution compared to its classical counterpart, and allows a quantum computer to be programmed by fuzzy linguistic rules. The proposed inference engine was implemented using a quantum algorithm design scheme based on the oracle notion. This scheme allows the modeling of a fuzzy rule-based system as a Boolean function, the oracle, which is able to reconstruct the relationships between the antecedent and consequent parts of fuzzy rules, and can be efficiently computed on a quantum computer. The suitability of the proposed quantum algorithm for use as a fuzzy inference engine was tested in a typical benchmark scenario, such as that provided by inverted pendulum control.

Recently, Variational Quantum Circuits (VQCs) are attracting considerable attention among quantum algorithms thanks to their robustness to the noise characterizing the current quantum devices. In detail, VQCs involve parameterized quan-tum circuits to be trained by means of a classical optimizer that makes queries to the quantum device. VQCs play a key role in several applications including quantum classifiers where the Hilbert space is used as feature space. Currently, the most used classical optimizer to learn V QCs is the gradient descent method. However, the so-called barren plateaus issue causes gradients of cost functions to become exceedingly small as the dimension of the classification problem is increased. As consequence, gradient descent method could be not efficient in real-world classification problems. This paper proposes to apply Genetic Algorithms (GAs) to train VQCs used as quantum classifiers. As shown in the experiments, the application of GAs results in accurate solutions obtained with a reduced number of queries to quantum devices.

Genetic Algorithms (GAs) are optimization methods that search near-optimal solutions by applying well-known operations such as selection, crossover and mutation. In particular, crossover and mutation are aimed at creating new solutions from selected parents with the goal of discovering better and better solutions in the search space. In literature, several approaches have been defined to create new solutions from the mating pool to try to improve the performance of genetic optimization. In this paper, the literature is enriched by introducing a new mating operator that harnesses the stochastic nature of quantum computation to evolve individuals in a classical genetic workflow. This new approach, named Quantum Mating Operator, acts as a multi-parent operator that identifies alleles' frequency patterns from a collection of individuals selected by means of conventional selection operators, and encodes them through a quantum state. This state is successively mutated and measured to generate a new classical chromosome. As shown by experimental results, GAs equipped with the proposed operator outperform those equipped with traditional crossover and mutation operators when used to solve well-known benchmark functions.

The Support Vector Machine (SVM) is a well-known supervised machine learning approach aimed at facing classification problems. Thanks to the exploitation of kernel functions, SVM succeeds to address also complex classification tasks involving non-linearly separable data. However, there are limitations to its success when the feature space becomes large, and the kernel functions become computationally expensive to estimate. Quantum-enhanced Support Vector Machine (QSVM) utilizes the properties of quantum computers to obtain an exponential speed up with respect to the conventional SVM by using the quantum state space as the feature space. Unfortunately, to benefit from these advantages, currently, it is necessary to have a background knowledge about quantum computing concepts and specific language skills. The goal of this paper is to introduce a web-based tool to make simple to all researchers, coming from several and heterogeneous scientific backgrounds, the application of QSVM to different real-world classification problems. To achieve this goal, the presented web tool involves data preprocessing techniques and a user-friendly interface. The suitability of the presented web tool is shown in an experimental session where QSVM is applied to solve well-known classification tasks.

Noisy Intermediate Scale Quantum (NISQ) devices are expected to demonstrate the real potential of quantum computing in solving hard problems. However, quantum noise that characterizes this kind of devices still remains an obstacle for their practical exploitation in real world scenarios. As a consequence, there is a strong emergence for error correction techniques aimed at making NISQ devices stable and fully operative. Unfortunately, current approaches for quantum error correction are prohibitive for NISQ devices because of the enormous multiplicative cost in resources that they require. For this reason, so-called quantum error mitigation methods are emerging as alternative approaches able to attenuate the quantum error as much as possible, without requiring a strong additional computational effort. Among the most error-prone operations, there is surely the quantum measurement. Conventionally, mitigation methods for quantum measurement error compute a so-called mitigation matrix capable of correcting results outputted by a quantum processor. In this paper, a new measurement error mitigation approach based on genetic algorithms is proposed to learn an appropriate mitigation matrix. As shown in the experimental session, the proposed measurement error mitigation method is comparable with or better than a conventional algebraic approach in terms of the Hellinger fidelity.

This paper introduces a study on fuzzy-based approaches aimed at addressing a crucial task in quantum computation: the evaluation of the similarity between quantum states. A quantum state is a mathematical entity that provides a probability distribution for the outcomes of each possible measurement of a quantum algorithm. Because quantum computers are still characterized by high noise in computation, output quantum states generated by quantum algorithms could be very far to be close to the ideal output quantum state computed by a noiseless quantum computer. As a consequence, there is a strong emergence for measures capable of assessing the similarity level of two quantum states, one ideal and the other real, to infer the quality of a quantum device in performing precise calculations and design appropriate quantum error correction schemes. This research proves that fuzzy methods are fully suitable to face this crucial challenge in a pioneering scenario such as that of quantum computing, as proved by their application on well-known quantum algorithms, such as Bernstein-Vazirani and Grover's algorithm.

Recently, Quantum Computing is entered in the so-called Noisy Intermediate-Scale Quantum (NISQ) era, where devices characterized by a few number of qubits are potentially able to overcome classical computers in performing specific tasks. However, noise in quantum operators still limits the size of quantum circuits that can be run in a reliable way. Consequently, there is a strong need for error mitigation approaches aimed at increasing reliability in quantum computation and making this paradigm really useful and productive in real world applications. In this paper, a fuzzy method, such as Fuzzy C-Means (FCM) clustering, has been used, for the very first time, to support the identification of matrices for error mitigation in quantum measurement. As shown in experiments, mitigation matrices identified with the support of FCM are able to strongly reduce error in computation when compared to mitigation matrices conventionally identified, like those used by IBM in its quantum library named Qiskit.

Quantum computing is a fascinating research area which promises a revolution in computing performance. Since the launch of the IBM Quantum Experience project in 2016, the research activities in this area are strongly increased. This project provides the public access to quantum processors composed of superconducting physical computing elements known as qubits. Unfortunately, qubits are sensitive to noise and, for this reason, quantum computation can be affected by errors. As a consequence, there is a strong emergence for so- called quantum error mitigation methods aimed to attenuate the quantum error as much as possible, without requiring a strong additional computational effort. Among the most error- prone operations, there is surely the quantum measurement. Conventionally, mitigation methods for quantum measurement error compute a so-called mitigation matrix capable of correcting results outputted by a quantum processor. In this paper, a new measurement error mitigation approach based on genetic algorithms whose fitness function uses Bhattacharyya distance is proposed to learn an appropriate mitigation matrix. As shown in the experimental session, the proposed measurement error mitigation method outperforms the traditional approach.

Bloodstain Pattern Analysis (BPA) is used by forensic officers to analyse bloodstains left at crime scenes. As a consequence, it has a crucial role in the investigations of bloody crimes. Currently, BPA activities are performed manually by leading to a slow and potentially imprecise analysis of the crime scenes. In order to overcome these issues, recently, some software tools have been developed in order to support forensic investigators in performing BPA activities in a more objective and fast way. However, only few approaches in literature use artificial intelligence methodologies to achieve this goal. Starting from this consideration, this paper presents a new intelligent tool based, for the first time, on the Density-Based Spatial Clustering of Application with Noise (DBSCAN) for supporting BPA activities. As shown in a case study, the proposed intelligent tool produces results that match with the ground truth.

Supervised learning methods aimed at performing precise predictions by learning from labeled training data. Unfortunately, training data can contain noisy or wrong information, specially when they come from real-world applications. In this scenario, applying a so-called training set selection procedure on data can lead to improve the performance of the supervised learning methods used for classification or regression tasks. In literature, several training set selection techniques have been proposed, but, to the best of our knowledge, few software tools implement this procedure. Moreover, all of them require programming capabilities or software package installation what makes their use difficult for people without specific computer skills. This paper proposes the first web-based tool, named TSSweb, for performing an accurate selection of the training instances. Thanks to its web nature, TSSweb enables all researchers, coming from several and heterogeneous scientific backgrounds, to reduce own datasets so as to improve their analysis and reduce the execution time of their supervised learning models. As shown in the experimental session, TSSweb produces reduced datasets with a good quality as well as being user-friendly.

One of the main issues in machine learning is related to the quality of data used to efficiently train statistical models for classification/regression tasks. Among these issues, the presence of missing values in data sets is particularly prone in affecting the accuracy performance of learning methods. As a consequence there is a strong emergence of software tools aimed at supporting machine learning users in "filling-in" their data sets before inputting them to training algorithms. This paper bridges this gap by introducing a web-based tool for MIssing DAta imputation (MIDA) based on a novel supervised learning method, namely Generalized Boosted Incremental Non Parametric Imputation algorithm (G-BINPI), able to address the missing values issue in scenarios where a "missing at random" assumption occurs. The proposed approach enables machine learning users to remotely imputing their data sets by means of an intuitive graphical user interface. As highlighted in the experimental section, the proposed approach yields better performance than conventional approaches for missing data imputation on different benchmark data sets.

Electroencephalography (EEG) headsets are wearable computing devices capable of recording electrical activity of the brain. These devices play a key role in the Brain-Computer Interfaces (BCIs) systems, i.e., systems capable of acquiring, processing and classifying EEG signals in order to control external devices such as wireless prosthetics. In spite of their crucial role, the current EEG headsets are very uncomfortable being composed of many wet electrodes. Hence, single-channel BCIs with dry electrodes are emerging like wearable devices more accepted by users. Unfortunately, this kind of device typically provides weaker and noisier signal that makes more challenging the classification task. This work is aimed at improving the quality of the classification of EEG signals, and in particular of Steady-State Visual Evoked Potentials (SSVEP), captured by single-channel EEG devices by using an evolutionary algorithm-based optimized version of Support Vector Machine (SVM). As shown by experimental results, the proposed approach improves on the state-of-the-art methods in terms of accuracy.

Sponsored by the IEEE Computational Intelligence Society, IEEE Std 1855 is aimed at defining a standard language, named Fuzzy Markup Language, capable of modeling fuzzy systems without considering hardware/software constraints and enabling, in this way, fuzzy systems sharing. In order to provide runnable fuzzy systems in accordance with IEEE Std 1855, recently, a Java library, named JFML, has been developed. Unfortunately, in spite of its benefits, JFML enables the modeling of fuzzy systems only for Java programmers. In order to overcome this drawback, this paper introduces a new open source fuzzy system software capable of enabling the modeling of fuzzy systems in accordance to IEEE Std 1855 enhancing JFML through a visual environment based on graphical user interfaces. Hence, VisualJFML represents a remarkable contribution to the literature since it allows modelling sharable fuzzy systems also to designers without programming skills. The user-friendly graphical interface provided by VisualJFML is shown by means of a case study dealing with the Iris classification problem.

Steady-State Visual Evoked Potentials (SSVEPs) are electroencephalography (EEG) signals which, recently, have attracted a notable interest in the field of Brain Computer Interfaces (BCIs) due to their little training requirement. Similar to other EEG signals, SSVEPs are captured by means of EEG devices characterized by multiple wet electrodes. Unfortunately, these EEG devices are very uncomfortable for users to be worn. As a consequence, there is a strong interest in developing more comfortable single-channel EEG devices equipped with dry sensors. However, if, on one hand, these innovative EEG devices are less invasive and more simple to be used by an average user, on the other hand, the exploitation of a single dry sensor leads to the collection of a weaker and noisy signal which is more difficult to be opportunely processed and classified. The aim of this paper is to improve the performance of signal classification tasks of SSVEPs captured by single-channel EEG devices with dry sensors by logistic regression. As shown by experimental results, the proposed approach improves the state-of-the-art methods in terms of accuracy.

Mobile devices are always more diffused in the last years, allowing the users to perform several tasks: communication, web surfing, requiring web services. Given the high amount of sensitive data and operations related to these tasks, securing the mobile devices is becoming a very critical issue. As matter of the fact, malware attacks are on the rise and new mobile malware are continually generated with the aim of stealing private data and performing illegal activities. Since this new malware is mainly obtained by reusing existing malicious code, malware detection is supported by the study and the tracking of the mobile malware phylogeny. This paper proposes a malware phylogeny model obtained by a declarative Process Mining (PM) approach from the analysis of some running malware applications. The main idea is that the set of relations and recurring execution patterns among the syscalls of a running malware application can be modeled to obtain a malware fingerprint. The malware fingerprints are compared and classified by using a fuzzy clustering algorithm to recover the malware phylogeny map of all the considered malware families. The evaluation of the proposed approach is performed on a dataset of more than 4,000 infected applications across 39 malware families obtaining very encouraging results.

Fuzzy Rule-Based Systems (FRBSs) are powerful tools for handling uncertainty in complex real world problems. Unfortunately, designing an optimal set of fuzzy membership functions and rules is not an easy task. The complexity of this task increases furthermore when the design involves type-2 fuzzy memberships because of the higher number of parameters to identify. For this reason, researches have been carried out to automatically extract the most suitable parameters and the hidden rules for type-2 FRBSs by using historical data. However, very few approaches exist based on a hybrid optimization between global and local search. This paper is aimed at overcoming this gap by proposing to automatically generate type-2 FRBSs through the so-called memetic algorithms. As shown in the experiments, the exploitation of memetic algorithms leads to generate type-2 FRBSs more accurate than those obtained by well-known approaches in literature.

This paper introduces the first implementation of fuzzy rule-based systems for quantum computers. Quantum computers are incredibly powerful machines that use principles of quantum mechanics, such as superposition and entanglement, to process information in a more efficient way than classical computers. Currently, several companies and academic research groups are focusing their efforts on the development of quantum algorithms for artificial intelligence but, so far, no significant contribution in the area of the fuzzy inference for quantum machines has been proposed. This article bridges this gap by introducing a quantum implementation of a fuzzy rule-based system through a synergistic usage of the well-known Grover's algorithm and look-up tables. As shown by a case study, the behaviour of the quantum implementation of the fuzzy rule-based system is comparable to its classical version, but with a quadratic asymptotic speed-up than a fuzzy rule-based system implemented by a classical look-up table.

Fuzzy Logic Systems have been successfully used in a wide range of real-world problems. They can include a priori expert knowledge and represent systems for which it is not possible to obtain a mathematical model. The standard IEEE Std 1855™-2016 was established to provide the fuzzy community with a unique and well-defined tool allowing a fuzzy logic system design completely independent from the specific hardware/software. Recently, the library Java Fuzzy Markup Language (JFML) offers a complete implementation of the standard, however, the actual version of the JFML does not support the development of fuzzy inference systems on specific types of hardware. The aim of this paper is to develop an interoperability module to design and running FLS for embedded systems in JFML, concretely for Arduino boards. In addition, a communication protocol between JFML and Arduino boards is also defined, removing the limited computing capacity usually offered by embedded systems. A case study with a wall-following fuzzy controller to manage a mobile robotic in two environments is developed in order to illustrate the potential of the new interoperability module.

In the last years, the growing adoption of cloud-based multi-tiers systems has strongly increased the levels of resource sharing among companies, improving the enterprise efficiency, thanks to a refined business dynamism and a rapid decrease in costs. However, in spite of their advantages, this new business model highlights the emergence of new computational approaches aimed at the distribution and the optimization of resources sharing along so-called multi-tenants system, i.e., cloud-based architecture where a single instance of software runs on a single server and serves multiple companies (tenants). This paper faces this challenging gap by proposing an auto-scaling cloud computing multi-tenancy architecture where process mining and fuzzy-based load-balancing systems synergistically interact to provide an improved and optimized resource management distribution. A case study is carried out to show the proposed architecture in operation.

Fuzzy Systems are an efficient instrument to create efficient and transparent models of the behavior of complex dynamic systems such as autonomous humanoid robots. The human interpretability of these models is particularly significant when it is applied to the cognitive robotics research, in which the models are designed to study the behaviors and produce a better understanding of the underlying processes of the cognitive development. From this research point of view, this paper presents a comparative study on training fuzzy based system to control the autonomous navigation and task execution of a humanoid robot controlled in a soccer scenario. Examples of sensor data are collected via a computer simulation, then we compare the performance of several fuzzy algorithms able to learn and optimize the humanoid robot's actions from the data.

IYEEE Std 1855 is the first IEEE standard technology developed in the area of fuzzy logic. Its main characteristic is the interoperability, a design feature that enables system designers to develop fuzzy inference engines without taking into account the hardware/software constraints imposed by the specific architecture on which the system will be deployed. Thanks to this feature, a fuzzy system can be integrated into different types of architectures without any need to carry out porting strategies. This feature is particularly crucial in the area of embedded systems where, for each kind of device, a variety of applications, communication protocols, software libraries and programming tools, exists. In this context, ArduinoTMtechnology represents one of the most popular architectures, thanks to its ease of development and prototyping. This paper shows how the native extendability feature of IEEE Std 1855 enables the design of a fuzzy rule-based systems in fully interoperable fashion on ArduinoTMarchitectures and, as a consequence, allows designers to focus on fuzzy concepts, without any need to consider the hardware/software details related to the specific ArduinoTMsystem.

This paper explores the applications of fuzzy logic inference systems as an instrument to perform linguistic analysis in the domain of prosodic prominence. Understanding how acoustic features interact to make a linguistic unit be perceived as more relevant than the surrounding ones is generally needed to study the cognitive processes needed for speech understanding. It also has technological applications in the field of speech recognition and synthesis. We present a first experiment to show how fuzzy inference systems, being characterised by their capability to provide detailed insight about the models obtained through supervised learning can help investigate the complex relationships among acoustic features linked to prominence perception.

A robotic system that is designed to coexist with humans has to adapt its behavioral and social interaction parameters not only with respect to the task it is supposed to accomplish, but also with respect to the human being it is interacting with by profiling her habits, preferences, and personality. This is particularly relevant in the domain of assistive robotics where the behavioral adaptability has been shown to enhance the users' acceptability of a robot. In this work, we propose a neuro-fuzzy-Bayesian system able to adapt the robot proxemics behavior with respect to the human users' personality and the action she is currently performing. The user's personality is evaluated according to the Big-Five factors model and the activity recognition is obtained by classifying data from a wearable device through the use of a Bayesian Network classifier. As shown by a statistical study, the proposed framework is capable of computing the most appropriate robot proxemics behavior in order to improve human feeling in interacting with artificial agents, such as robots.

Optimal Reactive Power Flow (ORPF) is one of the most fundamental and widely used tools in modern smart grids for improving the power system performance in terms of both operation cost and quality of service. This result is obtained by identifying the optimal set points of the voltage controllers, which improve the bus voltage profiles, minimise the regulation costs, and satisfy the power equipment constraints. In this context, the application of traditional optimisation methods is often complicated by the presence of multiple local minima, which could affect the performance and the convergence of iterative methods. To address this issue the adoption of evolutionary based paradigms has been recognised as one of the most promising enabling methodologies. Armed with such a vision, in this paper a novel solution method based on Search Group Algorithm (SGA) is proposed to solve the ORPF problem. As demonstrated by detailed experimental studies, which have been performed on IEEE 57-bus and 118-bus test power systems, the proposed evolutionary approach outperforms state-of-the-art evolutionary algorithms.

Memetic Algorithms (MAs) are a family of metaheuristics which combines global search approaches with local search techniques. Thanks to this hybridization, MAs have emerged as a powerful tool for tackling hard optimization problems in a more efficient way than their `no hybrid' counterparts. However, despite the success of MAs, it is still vital to note that the design of MAs raises a number of important issues which must be addressed to achieve a suitable hybridization of global and local search approaches resulting in a so-called Competent Memetic Algorithm (CMA). Currently, no software tools exist to automatically design CMAs and, as a consequence, all the hybridization choices are responsibility of human designers, which suffer from slowness and propensity to make mistakes. In order to bridge this design gap, this paper introduces jMeme, a Java library enabling an automatic development of CMAs so as to free practitioners of the responsibility for the design choices. As shown by experiments involving well-known benchmark functions, the competent design of MAs performed by jMeme yields better performance than MAs designed conventionally.

The k-nearest neighbor (k-NN) algorithm is one of the most well-known supervised classifiers due to its ease of use and good performance. However, in spite of its popularity, k-NN suffers from some drawbacks such as high computational complexity, high storage requirements, and low noise tolerance. Prototype selection is a successful technique aimed at addressing aforementioned issues by reducing the size of training datasets without deprecating, but improving, the classification accuracy. Recently, evolutionary algorithms have been successfully applied to the optimisation of accuracy and size of reduction of prototype selection because of their innate exploration and exploitation capabilities in visiting the space of solutions of a problem. However, so far, all the evolutionary approaches for prototype selection are based on a so-called multi-objective “a priori” technique, where multiple objectives are aggregated together into a single objective through a weighted combination. This paper proposes to apply, for the first time, an “a posteriori” algorithm, namely SPEA2, to prototype selection problem in order to explicitly deal with both objectives and offer a better trade-off between classification and reduction performance. As shown in the experimental section, the application of SPEA2 allows to hold high accuracy in nearest neighbour classification with a significant reduction of training data thanks to the discovery of higher quality solutions than those detected by a conventional “a priori” approach.

Fuzzy Cognitive Maps (FCMs) are a very simple and powerful technique for simulation and analysis of dynamic systems. In spite of their wide applicability in different domain areas, the manual development of FCMs suffers from several drawbacks such as the human difficulty to deal with systems characterised by a large number of variables. Therefore, several evolutionary learning approaches aimed at automatically building FCM models by using historical data have been developed over years. Nevertheless, there is no a formal and complete comparison able to evaluate the performance of evolutionary algorithms in learning FCMs. Consequently, the goal of this paper is to bridge this experimental gap by performing a multiple statistical procedure able to compare the best known nature-inspired algorithms-based learning methods for FCM models.

Inferring Gene Regulatory Networks (GRNs) from expression data is one of the most challenging topic in computational biology. Indeed, the reasoning about GRN behaviours is a crucial biological task useful to provide an significant support for the identification of genetic diseases and the estimation of the effects of medications. Over years, several approaches have been applied to infer GRNs, most of them are based on deterministic and crisp-based algorithms. However, the intrinsic imprecise nature of the gene regulation makes these approaches as inefficient and characterized by a low accuracy. Starting from this consideration, in this work, we propose to use Fuzzy Cognitive Maps to model the complex behaviour of GRNs and to learn FCMs models of GRNs by means of an innovative evolutionary algorithm: the Big Bang-Big Crunch algorithm. As shown through a statistical comparison, the proposed approach outperforms other evolutionary learning methods in inferring GRNs representing, as a consequence, a breakthrough approach in this fascinating and challenging domain.

Bloodstain pattern analysis (BPA) is a forensic discipline that plays a key role in performing a reconstruction of the crime scene related to blood shedding events. By studying the distribution, size and shape of bloodstains, BPA supports worldwide investigation agencies (US FBI, Italian Carabinieri and so on) in identifying the dynamics of a certain act of violence and evaluating the credibility of statements provided by a witness, a victim, or a suspect. However, in spite of its importance, this forensic discipline is still mainly based on manual approaches, making the analysis of a crime scene long, tedious and potentially imperfect. This paper is aimed at presenting a proposal for a robotic framework to automate the BPA in all its steps. In particular, the robotic framework is composed of an Unmanned Aerial Vehicle (UAV) capable of navigating the crime scene, detecting bloodstains, computing the points of origin and preparing a technical report describing the bloody event.

Fuzzy cognitive maps (FCMs) form an important class of models for describing and simulating the behavior of dynamic systems through causal reasoning. Owing to their abilities to make the symbolic knowledge processing simple and transparent, FCMs have been successfully used to model the behavior of complex systems originating from numerous application areas, such as economy, politics, medicine, and engineering. However, the design of FCMs necessarily involves domain experts to develop a graph-based model composed of a collection of system's concepts and causal relationships among them. Consequently, since humans exhibit an intrinsic factor of subjectivity and are only able to efficiently develop small-size graph-based models, there is a legitimate need to devise methods capable of automatically learning FCM models from data. This research addresses this need by introducing a competent memetic algorithm to generate FCM models from available historical data, with no human intervention. Extensive benchmarking tests performed on both synthetic and real-world data quantify the performance of the competent memetic method and emphasize its suitability over the models obtained by conventional and noncompetent hybrid evolutionary approaches in terms of accuracy, approximation ability, and convergence speed. Moreover, the proposed approach is shown to be scalable due to its capability to efficiently learn high-dimensional FCM models.

In recent years, several ontology-based systems have been developed for data integration purposes. The principal task of these systems is to accomplish an ontology alignment process capable of matching two ontologies used for modeling heterogeneous data sources. Unfortunately, in order to perform an efficient ontology alignment, it is necessary to address a nested issue known as ontology meta-matching problem consisting in appropriately setting some regulating parameters. Over years, evolutionary algorithms are appeared to be the most suitable methodology to address this problem. However, almost all of existing approaches work with a single function to be optimized even though a possible solution for the ontology meta-matching problem can be viewed as a compromise among different objectives. Therefore, approaches based on multi-objective optimization are emerging as techniques more efficient than conventional evolutionary algorithms in solving the meta-matching problem. The aim of this paper is to perform a systematic comparison among well-known multi-objective Evolutionary Algorithms (EAs) in order to study their effects in solving the meta-matching problem. As shown through computational experiments, among the compared multi-objective EAs, OMOPSO statistically provides the best performance in terms of the well-known measures such as hypervolume, Δ index and coverage of two sets.

During the last years, electronic markets (e-markets) are emerging as a new idea of economy, where trade transactions can be performed by buyers and sellers even if they are separated by geographic boundaries, time differences or distance barriers. Unfortunately, in this on-line trade environment, the probability of large-scale fraud and deceit is higher than traditional commerce because of the lack of face-to-face communications. For this reason, reputation systems, which enable to assess the trustiness level of transacting parties, are becoming a fundamental component of any current e-market portal. In this paper, we propose a new fuzzy logic based reputation system capable of efficiently assessing transacting parties through the exploitation of 1) a fuzzy trust model which takes into account a set of metrics reflecting the trust human perception both on seller and buyer side and, furthermore, it does not miss to consider past transactions; 2) a fuzzy based reputation aggregation taking into account credibility concept to discriminate false trust values. As shown by performed experiments, the proposed reputation system yields better performance than that used by one of the most known e-markets, eBay®.

The introduction of Fuzzy Markup Language (FML) in 2004 has initiated an important trend in Computational Intelligence research: the application of new web technologies to create more flexible and hardware independent environment for deploying "fuzzy ideas". FML allows researchers and engineers to focus on problem solving activities bypassing additional difficulties related to programming or physical equipment constraints. From that moment on, many researches have been using FML and other XML-based languages for modeling and developing fuzzy systems. However, in spite of their hardware interoperability, XML languages are able to model a fuzzy system in static way and, consequently, they do not provide any support for modelling "evolving" and temporal-based fuzzy systems, as such Timed Automata based Fuzzy Controllers. To address this deficiency, this paper introduces an extension of FML called FMLScript. It is based on a scripting language concept and allows for modeling XML-based systems that can dynamically modify their configurations. As the consequence, a better expressive power can be achieved when compared with static modelling approaches. This is shown in a case study involving a smart grid control.

Ontology alignment systems are software tools aimed at producing a set of correspondences, called alignment, between two heterogeneous ontologies in order to bring them in a mutual agreement. Performing this task is an essential step to allow the exchange of information between people, organizations and web applications using ontologies for representing their view of the world. Currently, in spite of several ontology alignment systems have been developed, there is no a robust solution that seems capable of producing alignments with the same high quality on different alignment task instances. Mainly, this weakness of ontology alignment systems is due to the dependence of their behavior on a set of specific instance parameters. This work proposes to improve performance of a well-known memetic algorithm based ontology alignment system by adaptively regulating its specific instance parameters through a FML-based fuzzy tuning. The validity of our proposal is shown by aligning ontologies belonging to two well-known OAEI datasets and by performing a Wilcoxon's signed rank test which highlights that our proposal statistically outperforms its not fuzzy adaptive counterpart.

Achieving semantic interoperability is an essential task for all distributed and open knowledge based systems. Currently, the best technology recognized for fulfilling this complex task is represented by ontologies. Unfortunately, in turn, the power of ontological representation is reduced by the semantic heterogeneity problem which affects two ontologies when they are characterized by terminological and conceptual discrepancies. The most solid solution to overcome this problem is to perform an ontology alignment process capable of leading two heterogeneous ontologies into a mutual agreement by detecting a set of correspondences between them. All ontology alignment processes based on evolutionary approaches developed so far perform an evaluation of the produced alignments based on multi-objectives "a priori" approaches. This paper proposes to apply NSGA II to ontology alignment problem in order to overcome the well-known drawbacks of "a priori" methods. As shown in the experimental section, the application of NSGA II allows to improving semantic interoperability by finding high quality solutions that are not detected by "a priori" approaches.

Interoperability is a key problem in agent-based systems where different interacting computational entities negotiate to achieve a common goal. In last years, this interoperability issue has been faced by exploiting the concept of ontology that enables a single agent to model its knowledge by means of a semantic description of a domain of interest. However, ontology ability to enable a full interoperability can be limited by the so-called semantic heterogeneity problem which arises when some discrepancies exist among ontologies modeling the knowledge related to different agents. As consequence, in order to enable an effective knowledge exchange, an ontology alignment process is necessary to lead proprietary ontologies to a mutual agreement. Recently, some studies have successfully investigated the suitability of memetic algorithms to solve this complex task. However, memetic algorithms are influenced by some design issues arising from the different choices that can be taken to implement them. The aim of this paper is to compare the performances yielded by different memetic ontology alignment systems in order to individuate the most suitable hybrid evolutionary approach which enables a strong agent interoperability. The comparison among the considered approaches is performed by applying a statistical multiple comparison procedure on a collection of ontologies belonging to the well-known Ontology Alignment Evaluation Initiative (OAEI) benchmarks.

Ontologies are recognized as a fundamental component for enabling interoperability across heterogeneous systems and applications. Indeed, they try to fit a common understanding of concepts in a particular domain of interest to support the exchange of information among people, artificial agents, and distributed applications. Unfortunately, because of human subjectivity, various ontologies related to the same application domain may use different terms for the same meaning or may use the same term to mean different things, raising the so-called heterogeneity problem. The ontology alignment process tries to solve this semantic gap by individuating a collection of similar entities belonging to different ontologies and enabling a full comprehension among different actors involved in a given knowledge exchanging. However, the complexity of the alignment task, especially for large ontologies, requires an automated and effective support for computing high-quality alignments. The aim of this paper is to propose a memetic algorithm to perform an efficient matching process capable of computing a suboptimal alignment between two ontologies. As shown by experiments, the memetic approach is more suitable for ontology alignment problem than a classical evolutionary technique such as genetic algorithms. © 2012 Wiley Periodicals, Inc.

Born primarily as means to model knowledge, ontologies have successfully been exploited to enable knowledge exchange among people, organizations and software agents. However, because of strong subjectivity of ontology modeling, a matching process is necessary in order to lead ontologies into mutual agreement and obtain the relative alignment, i.e., the set of correspondences among them. The aim of this paper is to propose a memetic algorithm to perform an automatic matching process capable of computing a suboptimal alignment between two ontologies. To achieve this aim, the ontology alignment problem has been formulated as a minimum optimization problem characterized by an objective function depending on a fuzzy similarity. As shown in the performed experiments, the memetic approach results more suitable for ontology alignment problem than other evolutionary techniques such as genetic algorithms.

The rapid changes in modern knowledge, due to exponential growth of information sources, are complicating learners' activity. For this reason, novel approaches are necessary to obtain suitable learning solutions able to generate efficient, personalized and flexible learning experiences. From this point of view, the use of different cooperative intelligent agents can be exploited to analyze learner's preferences and generate high quality learning presentations which provide attractive learning solutions. In particular, to achieve this goal this paper exploits an ontological representation of the learning environment and an adaptive memetic algorithm based on a cooperative multi-agent framework. In this framework different agents analyze the e-learning instance and solve it in a parallel way, cooperating among them. This cooperation is performed by jointly exploiting data mining, via fuzzy decision trees, together with a decision making framework exploiting fuzzy methodologies. As will be shown in the experimental results section, this multi-agent strategy is capable of speeding up the convergence to high-quality personalized e-learning experiences.

In this paper we present the preliminary results of application of Fuzzy Markup Language (FML) to suspect a non-common disease. Under non-common diseases we understand rare diseases. From the broad point of view this problem belongs to the computer-assisted decision support in medical diagnostics and can be supported by fuzzy logic controllers. We can use conventional methods to diagnose a rare disease if it can be exhibited by outstanding symptoms. For example, there are several search machines and data banks that allow to find a rare disease clearly exhibited by a patient's symptoms/signs. But it is very difficult to diagnose a rare disease if it masks as a common disease. Diagnostic of rare diseases is connected with lack, uncertainty and imprecision of knowledge, medical mistake and even medical failure. Additionally, very often a common disease is also established with some degree of belief, thus, the expressions such as "it is possible that a patient has a particular disease" rather often present in the daily medical practice. It is clear that if we would know the common diseases, then deviations from them can be considered as a sign of non-common diseases. In this paper we investigate such deviations with the help of FML. We show how FML mechanism can be adjusted to suspect a rare disease, and discuss the appropriateness of the available operators.

The large-scale deployment of the Smart Grid paradigm will support the evolution of conventional electrical power systems toward active, flexible and self-healing web energy networks composed of distributed and cooperative energy resources. In a Smart Grid platform, the optimal coordination of distributed voltage controllers is one of the main issues to address. In this field, the application of traditional control paradigms has some disadvantages that could hinder their application in Smart Grids where the constant growth of grid complexity and the need for massive pervasion of Distribution Generation Systems (DGSs) require more scalable, more flexible control and regulation paradigms. To try and overcome these challenges, this paper proposes the concept of a decentralized non-hierarchical voltage regulation architecture based on intelligent and cooperative smart entities. The distributed voltage controllers employ traditional sensors to acquire local bus variables and mutually coupled oscillators to assess the main variables that characterize the operation of the global Smart Grid. These variables are then amalgamated by a novel fuzzy inference engine, named Timed Automata based Fuzzy Controllers, in order to identify proper control actions aimed at improving the grid voltage profile and reducing power losses.

Learning is a mechanism to acquire new knowledge and to enhance individual skills in industrial and academic environments. In particular, employing learning methods in an industrial context supports the overall business competitiveness in the new economy. Currently, the e-Learning systems provide a simple “digitalization” of the learning process where the focus is on the educational resources, which are only an input of the whole learning process, and on their presentation (delivery). Computational Intelligence methodologies can overcome current learning systems limitations attaining to personalize learning content and activities to specific preferences of the learner and to assist designers with computationally efficient methods to develop “in time” e-Learning environments. This paper shows how to achieve both results exploiting an ontological representation of learning environment and memetic approach of optimization, integrated into a cooperative distributed problem solving framework.

During the past several years, fuzzy control has emerged as one of the most suitable and efficient methods for designing and developing complex systems in environments characterized by high level of uncertainty and imprecision. Nowadays, this methodology is used to model systems in several applications domains which range from industrial machineries to financial decisions support systems. Nevertheless, in spite of the usefulness of fuzzy control, one of its drawbacks comes from the lack of the temporal concept that is crucial in many systems characterized from a discontinuous nonlinear behaviour. In particular, in its standard vision, fuzzy control is not able to represent Variable-Structure systems, i.e., systems that change their configuration (knowledge base) or their behaviour (rule base) over time. To overcome these drawbacks, this paper extends fuzzy control idea by considering a theory from formal languages: timed automata. This novel synergic approach achieves a twofold advantage by representing a system in qualitative and linguistic way and introducing a novel switching control concept able to maximize system's performances and robustness.