In teletraffic engineering, M/G/1 queues are pivotal for optimizing network performance and operational efficiency. Closed formulas like the mean-value Pollaczek-Khinchin (MV-PK) are highly valued by practitioners due to their simplicity, facilitating rapid and flexible system analyses. This formula aids in verifying the reliability of complex simulations concerning M/G/1 models, especially when applicable. The challenge intensifies when dimensioning M/G/1 systems with heavy-tailed service time distributions. Simulation of such queues becomes notably arduous, necessitating increased number of samples and longer simulation durations. Moreover, verification becomes more intricate as the MV-PK formula cannot be straightforwardly applied in such cases. This paper extends the MV-PK formula to heavy tailed service distributions with infinite/infinitesimal moments using Nonstandard Analysis. Additionally, it shows how recently introduced Bounded Algorithmic Numbers (BANs) enables numerical verification of the extended PK formula via discrete-event simulations of the M/G/1 queue, even when predicted delay values are infinite. The implemented approach tends to converge fast and exhibits remarkable numerical robustness. It adheres to the principle of “write once, run multiple times”, as the same source code for queue simulation handles both finite and infinite variance cases.

Machine Learning approaches have been successfully used for the creation of high-performance control components of cyber–physical systems, where the control dynamics result from the combination of many subsystems. However, these approaches may lack the trustworthiness required to guarantee their reliable application in a safety-critical context. In this paper, we propose a combination of interval arithmetic and theorem-proving verification techniques to analyze safety properties in closed-loop systems that embed neural network components. We show the application of the proposed approach to a model-predictive controller for autonomous driving comparing the neural network verification performance with other existing tools. The results show that open-loop neural network verification through interval arithmetic can outperform existing approaches proving properties with a smaller time overhead. Furthermore, we demonstrate the capability of combining the two approaches to construct a formal model of the network in higher-order logic of the controlled system in a closed-loop.

Abstract This work aims at reviewing the state of the art of the field of lexicographic multi/many-objective optimization. The discussion starts with a review of the literature, emphasizing the numerous application in the real life and the recent burst received by the advent of new computational frameworks which work well in such contexts, e.g., Grossone Methodology. Then the focus shifts on a new class of problems proposed and studied for the first time only recently: the priority-levels mixed-pareto-lexicographic multi-objective-problems (PL-MPL-MOPs). This class of programs preserves the original preference ordering of pure many-objective lexicographic optimization, but instantiates it over multi-objective problems rather than scalar ones. Interestingly, PL-MPL-MOPs seem to be very well qualified for modeling real world tasks, such as the design of either secure or fast vehicles. The work also describes the implementation of an evolutionary algorithm able to solve PL-MPL-MOPs, and reports its performance when compared against other popular optimizers.

Abstract Photonics-based neural networks promise to outperform electronic counterparts, accelerating neural network computations while reducing power consumption and footprint. However, these solutions suffer from physical layer constraints arising from the underlying analog photonic hardware, impacting the resolution of computations (in terms of effective number of bits), requiring the use of positive-valued inputs, and imposing limitations in the fan-in and in the size of convolutional kernels. To abstract these constraints, in this paper we introduce the concept of Photonic-Aware Neural Network (PANN) architectures, i.e., deep neural network models aware of the photonic hardware constraints. Then, we devise PANN training schemes resorting to quantization strategies aimed to obtain the required neural network parameters in the fixed-point domain, compliant with the limited resolution of the underlying hardware. We finally carry out extensive simulations exploiting PANNs in image classification tasks on well-known datasets (MNIST, Fashion-MNIST, and Cifar-10) with varying bitwidths (i.e., 2, 4, and 6 bits). We consider two kernel sizes and two pooling schemes for each PANN model, exploiting $$2\times 2$$ 2 × 2 and $$3\times 3$$ 3 × 3 convolutional kernels, and max and average pooling, the latter more amenable to an optical implementation. $$3\times 3$$ 3 × 3 kernels perform better than $$2\times 2$$ 2 × 2 counterparts, while max and average pooling provide comparable results, with the latter performing better on MNIST and Cifar-10. The accuracy degradation due to the photonic hardware constraints is quite limited, especially on MNIST and Fashion-MNIST, demonstrating the feasibility of PANN approaches on computer vision tasks.

Abstract With the arrival of the open-source RISC-V processor architecture, there is the chance to rethink Deep Neural Networks (DNNs) and information representation and processing. In this work, we will exploit the following ideas: i) reduce the number of bits needed to represent the weights of the DNNs using our recent findings and implementation of the posit number system, ii) exploit RISC-V vectorization as much as possible to speed up the format encoding/decoding, the evaluation of activations functions (using only arithmetic and logic operations, exploiting approximated formulas) and the computation of core DNNs matrix-vector operations. The comparison with the well-established architecture ARM Scalable Vector Extension is natural and challenging due to its closedness and mature nature. The results show how it is possible to vectorize posit operations on RISC-V, gaining a substantial speed-up on all the operations involved. Furthermore, the experimental outcomes highlight how the new architecture can catch up, in terms of performance, with the more mature ARM architecture. Towards this end, the present study is important because it anticipates the results that we expect to achieve when we will have an open RISC-V hardware co-processor capable to operate natively with posits.

This article concerns the study of mixed Pareto-lexicographic multiobjective optimization problems where the objectives must be partitioned in multiple priority levels (PLs). A PL is a group of objectives having the same importance in terms of optimization and subsequent decision making, while between PLs a lexicographic ordering exists. A naive approach would be to define a multilevel dominance relationship and apply a standard EMO/EMaO algorithm, but the concept does not conform to a stable optimization process as the resulting dominance relationship violates the transitive property needed to achieve consistent comparisons. To overcome this, we present a novel approach that merges a custom nondominance relation with theGrossonemethodology, a mathematical framework to handle infinite and infinitesimal quantities. The proposed method is implemented on a popular multiobjective optimization algorithm (NSGA-II), deriving a generalization of it called by us PL-NSGA-II. We also demonstrate the usability of our strategy by quantitatively comparing the results obtained by PL-NSGA-II against other priority and nonpriority-based approaches. Among the test cases, we include two real-world applications: one 10-objective aircraft design problem and one 3-objective crash safety vehicle design task. The obtained results show that PL-NSGA-II is more suited to solve lexicographical many-objective problems than the general purpose EMaO algorithms.

Nowadays, real-time applications are exploiting DNNs more and more for computer vision and image recognition tasks. Such kind of applications are posing strict constraints in terms of both fast and efficient information representation and processing. New formats for representing real numbers have been proposed and among them the Posit format appears to be very promising, providing means to implement fast approximated version of widely used activation functions in DNNs. Moreover, information processing performance are continuously improved thanks to advanced vectorized SIMD (single-instruction multiple-data) processor architectures and instructions like ARM SVE (Scalable Vector Extension). This paper explores both approaches (Posit-based implementation of activation functions and vectorized SIMD processor architectures) to obtain faster DNNs. The two proposed techniques are able to speed up both DNN training and inference steps.

Worldwide, four people die every minute as a consequence of illnesses and accidents at work. This considerable number makes occupational safety an important research area aimed at obtaining safer and safer workplaces. This paper presents a semi-supervised learning-aided evolutionary approach to improve occupational safety by classifying workers depending on their own risk perception for the task assigned. More in detail, a semi-supervised learning phase is carried out to initialize a good population of a non-dominated sorting genetic algorithm (NSGA-II). Each chromosome of the population represents a pair of classifiers: one determines a worker's risk perception with respect to a task, the other determines the level of caution of the same worker for the same task. Learning from constraints reinforces the initial training performance. The best Pareto-optimal solution to the problem is selected by means of the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS). The proposed framework was tested on real-world data gathered through a website purposely developed. Results showed a good performance of the obtained classifiers, thus validating the effectiveness of the proposed approach in supporting the decision-maker in critical job assignment problems, where risks are a serious threat to the workers' health.

This paper presents a method, based on classification techniques, for automatic detection and diagnosis of defects of rolling element bearings. The experimental data set consists of vibration signals recorded by four accelerometers on a mechanical device including rolling element bearings: the signals were collected both with all faultless bearings and after substituting one faultless bearing with an artificially damaged one. Four defects and, for one of them, three severity levels are considered. Classification accuracy higher than 99% was achieved in all the experiments performed on the vibration signals represented in the frequency domain, thus proving the high sensitivity of our method to different types of defects and to different degrees of fault severity. The degree of robustness of our method to noise is also assessed by analyzing how the classification performance varies with the signal-to-noise ratio and using statistical classifiers and neural networks.

The advent of relatively cheap general purpose graphics processing units (GPUs) is having a huge impact on scientific computing. This is opening the door to high performance fuzzy computing (HPFC) to the masses, due to the low cost and the possibility to have GPUs on desktop computers at home. Furthermore, even mobile HPFC seems imminent, provided that your laptop is equipped with a general purpose GPU. Very recently another innovation has occurred: the availability of libraries for high level GPU programming. By using them, the programmer avoids the necessity of having detailed knowledge of the GPU's hardware architecture. The availability of such facility on Matlab (today among the most used rapid prototyping software) is also opening the door to the rapid prototyping of high performance applications, in general, and to the rapid prototyping of HPFC, in particular. In this work we show how to speed up an existing Matlab software prototype (which computes spatial maps for supporting decision making during maritime operations) with little effort. By delegating the GPU Matlab library to take care of low level optimizations, we not only save time, but also build rapidly prototyped software that is portable over different GPU hardware types.

This paper presents a flexible approach to forecasting of energy production in solar photovoltaic (PV) installations, using time series analysis and neural networks. Our goal is to develop a one day-ahead forecasting model based on an artificial neural network with tapped delay lines. Despite some methods already exist for energy forecasting problems, the main novelty of our approach is the proposal of a tool for the technician of a PV installation to correctly configure the forecasting model according to the particular installation characteristics. The correct configuration takes into account the number of hidden neurons, the number of delay elements, and the training window width, i.e., the appropriate number of days, before the predicted day, employed for the training. The irradiation along with the sampling hour are used as input variables to predict the daily accumulated energy with a percentage error less than 5%.

This paper first reviews the state-of-the-art of fuzzy rule-based classifiers (FRBCs), then it discusses how to implement an FRBC under the Pattern Recognition Toolbox (PRTools), the de-facto standard toolbox for classification in Matlab. Such an implementation, called frbc, allows for a straightforward comparison of frbc with other classifiers already available under the PRTools. Furthermore, frbc can easily be used and combined with any other general-purpose function already available in PRTools. In this way, e.g., it becomes really easy to perform many types of feature selection, based on the accuracy achieved by frbc on the subset of features at hand. Another useful feature is the capability to export each FRBC generated by frbc as a standard Fuzzy Inference System (FIS) structure used within the Matlab Fuzzy Logic Toolbox (FLT): this allows comparisons/validations, visual inspection of the rule base, etc. In the experimental part we first assess the correctness of the implementation, by reproducing results existing in the literature. Then we show some examples of usage of frbc, combined with existing PRTools functions.

In this paper, we compare traditional classifiers, such as Linear and Quadratic Discriminant Classifiers and neural networks, with a one-class classifier, namely, convex hull. With reference to rolling element bearing diagnosis, we show that convex hull outperforms traditional classifiers in the classification of faults and different levels of fault severity not known during the training phase.

This paper shows an attempt to build a time variant cost-oriented classifier for two-class classification problems. Such classifier is based on a sliding window, and has been designed as an ensemble of Cost-Oriented Support Vector Machines (CO-SVMs). More precisely, we have integrated the Incremental/Decremental (ID) formulation of SVMs with the Cost-Oriented (CO) formulation, thus obtaining an ensemble of COID-SVMs. At each data arrival, the new pattern is classified by using a dynamic selection of the underlying COID-SVMs in the Receiver Operating Characteristic (ROC) space by means of the ROC convex hull method. Then, once the actual class label of the new pattern is known, the new data and the associated class label are used to perform an incremental learning by each COID-SVM. At the same time, each SVM is updated by performing the decremental learning of the data falling outside the sliding window. This allows to adapt the classification to time varying conditions. The methodology has been applied to the classification of oil spills at sea from remotely sensed optical images.

This paper presents a method, based on classification techniques, for automatic detection and diagnosis of defects of rolling element bearings. We used vibration signals recorded by four accelerometers on a mechanical device including rolling element bearings: the signals were collected both with all faultless bearings and after substituting one faultless bearing with an artificially damaged one. We considered four defects and, for one of them, three severity levels. In all the experiments performed on the vibration signals represented in the frequency domain we achieved a classification accuracy higher than 99%, thus proving the high sensitivity of our method to different types of defects and to different degrees of fault severity. We also assessed the degree of robustness of our method to noise by analyzing how the classification performance varies on variation of the signal-to-noise ratio and using statistical classifiers and neural networks. We achieved very good levels of robustness.

This paper presents a method, based on classification techniques, for automatically detecting and diagnosing various types of defects which may occur on a rolling element bearing. In the experiments we have used vibration signals coming from a mechanical device including more than ten rolling element bearings monitored by means of four accelerometers: the signals have been collected both with all faultless bearings and substituting one faultless bearing with an artificially damaged one: four different defects have been taken into account. The proposed technique considers all the aspects of classification: feature selection, different base classifiers (two statistical classifiers, namely LDC and QDC, and MLP neural networks) and classifier fusion. Experiments, performed on the vibration signals represented in the frequency domain, have shown that the proposed classification method is highly sensitive to different types of defects and to different severity degrees of the defects.

We have developed a multi-classifier system for automatic classification of pulmonary nodules in lung CT (Computed Tomography) images. The system consists of a set of independent modules, each emulating a radiologist of a team, and a further module aimed at appropriately combining theradiologists' opinions. In the experiments we obtained a sensitivity of 95% against a specificity of 91.33%, adopting a combiner based on the decisio.n templates technique.

Multiobjective genetic fuzzy systems (MGFSs) have proved to be very effective in classification, regression and control tasks. However, large scale problems still present open and challenging research issues. Making identification of fuzzy rules faster can enlarge the range of applications of MGFSs. In this work we first analyze the time complexity for both the identification and the evaluation of Takagi-Sugeno fuzzy rule-based systems. Then we introduce a simple but effective idea for fast identification of consequent parameters, although in an approximated, suboptimal manner. In the experimental part we first test the correctness of the predicted asymptotical time complexity. Then we show the benefits through an example of multiobjective genetic learning of compact and accurate fuzzy systems, in which we saved 71.3% of time on a 7 input problem.

In this paper, we propose an approach to complexity reduction of Mamdani-type fuzzy rule-based systems (FRBSs) based on removing logical redundancies. We first generate an FRBS from data by applying a simplified version of the well-known Wang and Mendel method. Then, we represent the FRBS as a multi-valued logic relation. Finally, we apply MVSIS, a tool for circuit minimization and simulation, to minimize the relation and consequently to reduce complexity of the associated FRBS. Unlike similar previous approaches proposed in the literature, the use of MVSIS let us deal with nondeterminism, that is, let us manage rules with the same antecedent but different consequents. To allow nondeterminism guarantees to achieve a higher (or at least not worse) complexity reduction than the one achievable from removing nondeterminism as soon as it appears. We apply our approach to six popular benchmarks. Results show a considerable complexity reduction associated only sporadically with consistent accuracy degradation. Moreover, quite surprisingly, the complexity reduction often comes together with an improvement in the classification accuracy.

In this paper, we propose a novel population- based multi-objective evolutionary algorithm (MOEA) for binary classifier optimization. The two objectives considered in the proposed MOEA are the false positive rate (FPR) and the true positive rate (TPR), which are the two measures used in the ROC analysis to compare different classifiers. The main feature of our MOEA is that the population evolves based on the properties of the convex hulls defined in the FPR-TPR space. We discuss the application of our MOEA to determine a set of fuzzy rule-based classifiers with different trade-offs between FPR and TPR in lung nodule detection from CT scans. We show how the Pareto front approximation generated by our MOEA is better than the one generated by NSGA-II, one of the most known and used population-based MOEAs.

An approach to select the most suitable fuzzy rule-based binary classifier to a specific application is proposed. First, an evolutionary three-objective optimization algorithm is applied to generate an approximation of a Pareto front composed of fuzzy rule-based binary classifiers with different trade-offs between accuracy and complexity. Accuracy is measured in terms of sensitivity and specificity, whereas complexity is computed as sum of the conditions which compose the antecedents of the rules included in the classifiers. Thus, low values of complexity correspond to fuzzy systems characterized by a low number of rules and a low number of input variables actually used in each rule. This ensures a high comprehensibility of the classifiers. Then, the most suitable classifier is selected by using the ROC convex hull method. We discuss the application of the proposed approach to generate a classifier for discriminating lung nodules from non-nodules in a computer aided diagnosis (CAD) system. Results obtained on a real data set extracted from lung CT images are also discussed

In this paper, we propose a fuzzy logic-based approach which exploits remotely sensed multispectral measurements of the reflected sunlight to estimate the concentration of optically active constituents of the sea water. The relation between the concentrations of interest and the subsurface reflectances is modeled by a set of fuzzy rules extracted automatically from the data through a two-step procedure. First, a compact initial rule base is generated by projecting onto the input variables the clusters produced by a fuzzy clustering algorithm. Then, a genetic algorithm is applied to optimize the rules. Appropriate constraints maintain the semantic properties of the initial model during the genetic evolution. Results of the application of the fuzzy model obtained from data simulated with an ocean color model over the channels of the Medium Resolution Imaging Spectrometer are shown and discussed.

In this paper, we present a system based on an Unsupervised Fuzzy Divisive Hierarchical Clustering (UFDHC) algorithm to determine a hierarchy of profiles of web site typical users from the web access log. These profiles can be extremely useful, for instance, to customize the web site, or to send personalized advertisements. After eliminating categories that have not been accessed by a significant percentage of users and removing the occasional users, the access log data are input to the UFDHC algorithm which clusters the users of the web site into a hierarchy of groups characterized by a set of common interests and represented by a prototype, which defines the profile of the group typical member. To show the effectiveness of our system, we describe how the profiles determined by the UFDHC algorithm from access log data collected along a period of 15 days allow classifying approximately 95% of the users defined by access log data collected during subsequent 60 days.

In this paper, we describe a fuzzy model for the estimation of sea water optically active constituents concentration in case II water. The model is extracted automatically from the data through a two step procedure. First, a fuzzy clustering algorithm is applied to identify a compact initial rule-based model. Then, the model is optimized by means of a genetic algorithm which tunes the rules so as to minimize the error between desired and predicted outputs. Appropriate constraints maintain the semantic properties of the initial model during the genetic evolution. The fuzzy model has been tailored to the multispectral data format of the MEdium Resolution Imaging Spectrometer (MERIS) on board the ESA-ENVISAT satellite launched in March 2002.