Computational Science and Its Applications – ICCSA 2022

Artificial neural networks (ANNs) have been extensively used for the description of problems arising from biological systems and for constructing neuromorphic computing models. The third generation of ANNs, namely, spiking neural networks (SNNs), inspired by biological neurons enable a more realistic mimicry of the human brain. A large class of the problems from these domains is characterized by the necessity to deal with the combination of neurons, spikes and synapses via integrate-and-fire neuron models. Motivated by important applications of the integrate-and-fire of neurons in neuromorphic computing for bio-medical studies, the main focus of the present work is on the analysis of the effects of additive and multiplicative types of random input currents together with a random refractory period on a leaky integrate-and-fire (LIF) synaptic conductance neuron model. Our analysis is carried out via Langevin stochastic dynamics in a numerical setting describing a cell membrane potential. We provide the details of the model, as well as representative numerical examples, and discuss the effects of noise on the time evolution of the membrane potential as well as the spiking activities of neurons in the LIF synaptic conductance model scrutinized here. Furthermore, our numerical results demonstrate that the presence of a random refractory period in the LIF synaptic conductance system may substantially influence an increased irregularity of spike trains of the output neuron.

Graph clustering and community detection are significant and actively developing topics in network science. Uncovering community structure can provide essential information about the underlying system. In this work, we consider two closely related graph clustering problems. One is the clique partitioning problem, and the other is the maximization of partition quality function called modularity. We are interested in the exact solution. However, both problems are NP-hard. Thus the computational complexity of any existing algorithm makes it impossible to solve the problems exactly for the networks larger than several hundreds of nodes. That is why even a small reduction of network size can significantly improve the speed of finding the solution to these problems. We propose a new method for reducing the network size that preserves the optimal partition in terms of modularity score or the clique partitioning objective function. Furthermore, we prove that the optimal partition of the reduced network has the same quality as the optimal partition of the initial network. We also address the cases where a previously proposed method could provide incorrect results. Finally, we evaluate our method by finding the optimal partitions for two sets of networks. Our results show that the proposed method reduces the network size by 40% on average, decreasing the computation time by about 54%.

A new finite-difference approximation of the two-dimensional parabolic equation is proposed in this paper. The specifics of the tropospheric radio-wave propagation problem are taken into account. Rational approximation of the discrete in both dimensions propagation operator is considered. The method of rational interpolation is used instead of local Padé approximation. The results of numerical modeling confirm the advantages of the proposed approach.

In many situations, the analysis of viscoelastic materials, like polymers, takes benefit from the introduction of fractional operators in the mathematical formalization. In addition, fractional differential models have been applied in a wide variety of fields, from biology to thermodynamics, from diffusion of information to dehydration/rehydration of food. Thus, a great interest is paid both in the analytical and in the numerical solution of fractional differential problems. The present paper considers a class of time-fractional diffusion problems with Dirichlet boundary conditions. Using Duhamel’s principle, the analytical solution is found. As usual in this context, the solution is given in series form and depends on the Mittag-Leffler function. We suggest a computational procedure to evaluate the solution with high accuracy, in a computing environment. Some test examples are presented both in the subdiffusion and in the superdiffusion case, to illustrate the behavior of the solution for different values of the fractional index. Test cases have been carried out in Matlab.

Towards the end of 2020, as people changed their usual behavior due to end of year festivities, increasing the frequency of meetings and the number of people who attended them, the COVID-19 local epidemic’s dynamic changed. Since the beginnings of this pandemic, we have been developing, calibrating and validating a local agent-based model (AbcSim) that can predict intensive care unit and deaths’ evolution from data contained in the state electronic medical records and sociological, climatic, health and geographic information from public sources. In addition, daily symptomatic and asymptomatic cases and other epidemiological variables of interest disaggregated by age group can be forecast. Through a set of Hidden Markov Models, AbcSim reproduces the transmission of the virus associated with the movements and activities of people in this city, considering the behavioral changes typical of local holidays. The calibration and validation were performed based on official data from La Rioja city in Argentina. With the results obtained, it was possible to demonstrate the usefulness of these models to predict possible outbreaks, so that decision-makers can implement the necessary policies to avoid the collapse of the health system.

The color-spanning problem for a given set of colored geometric objects finds a region that contains at least one object of each color. This paper studies the color-spanning problem for a given set of colored line segments $$\mathcal{L}$$ L lying in $${\mathbb {R}}^2$$ R 2 . For n line segments, each associated with any of m colors ( $$3\le m< n$$ 3 ≤ m < n ), we compute (i) one (resp. two congruent disjoint) color-spanning vertical strip (resp. strips) of minimum width in O(n) time after $$O(n\log n)$$ O ( n log n ) preprocessing task, (ii) one (resp. two congruent disjoint) color-spanning axis parallel square (resp. squares) of minimum perimeter in $$O(n(n-m))$$ O ( n ( n - m ) ) (resp. $$O(n^3)$$ O ( n 3 ) ) time. The same results also exist if the inputs to the aforesaid problems are n simple polygons or disks. Our algorithm improved the time complexity of finding the minimum perimeter color-spanning square problem for the segments [1] by $$\log n$$ log n factor. Finally, we show the $$\sqrt{2}$$ 2 approximation algorithm for computing one (resp. two congruent disjoint) color-spanning disk (resp. disks) of minimum radius for $$\mathcal L$$ L .

In this work we propose a novel and alternative interpretation of the SEIR model, typically used in epidemiology to describe the spread of a disease in a given population, to describe the diffusion of fake information on the web and the consequent truth re-affirmation. We describe the corresponding system of ordinary differential equations, giving a proper definition of the involved parameters and, through a local linearization of the system, we calculate the so-called stiffness ratio, i.e. the ratio between the real parts of the largest and smallest eigenvalues of the Jacobian matrix of the linearized problem. A large gap in the spectrum of such a Jacobian matrix (i.e., a large stiffness ratio) makes the underlying differential problem stiff. So, we study and analyze the stiffness index of the SEIR model and, through selected numerical examples on real datasets, we show that the more the model is stiff, the faster is the transit of fake information in a given population.

The privacy of the trajectories of indoor space users is just as important as that of the users of outdoor spaces. Many users of indoor spaces consider it very important to maintain the privacy of their movements within buildings and not reveal their visit to a certain room/cell inside buildings. In this paper, we propose a cloaking diversity approach for moving entities in indoor spaces. This new privacy-assurance approach uses the cloaking concept adapted with processing diversity trajectories to safeguard user privacy in an indoor space. Extensive simulations and evaluations have demonstrated that the proposed privacy approach algorithm performs well and at a low cost.

Performing data augmentation for mixed datasets remains an open challenge. We propose an adaptation of the Mixed Deep Gaussian Mixture Models (MDGMM) to generate such complex data. The MDGMM explicitly handles the different data types and learns a continuous latent representation of the data that captures their dependence structure and can be exploited to conduct data augmentation. We test the ability of our method to simulate crossings of variables that were rarely observed or unobserved during training. The performances are compared with recent competitors relying on Generative Adversarial Networks, Random Forest, Classification And Regression Trees, or Bayesian networks on the UCI Adult dataset.

This study proposes a numerical technique based on a hybrid of block-pulse functions and Chelyshkov polynomials to solve fractional delay differential equations. The Galerkin approach transforms the solution of fractional delay differential equations into a system of algebraic equations using the fractional operational matrix of integration for these hybrid functions. The suggested method’s accuracy and efficiency are demonstrated using numerical examples.

The paper considers the influence of the scattered component of the wavefield on the full waveform inversion results. Utilizing the one step of the quasi-Newton method, we demonstrate that scattered wavefields bring helpful information in the context of the solution of the inverse dynamical problem. Theoretically, usage of the scattered waves should increase the resolution of the method. For different scenarios, the contribution of scattered waves is investigated numerically. We investigate the influence of scattered components only by using singular value decomposition of the linearized inverse problem operator. The numerical experiments are performed for the well-known Marmousi2 model.

The multi-section robots also called variable geometry robots (VGT), are formed by different modules and have multiple degrees of freedom (DOF); These robots are a new class that can be defined as systems adaptable to different environments, unlike conventional robots, multi-section robots allow greater flexibility and adaptability to carry out tasks with restricted space conditions, their locomotion has a high degree of manipulation and dexterity in environments with difficult access and very closed spaces where maneuverability must be high, these characteristics are very similar to those exhibited by the movements of snakes, elephant trunks, and octopus tentacles, capabilities beyond them reach of traditional handlers of rigid link, multi-link robots can adapt their shape to navigate through complex environments. In this work, we show the implementation of the Lie Matrix Theory of the rigid movements of a body in a multi-link Robot so that through kinematics and with the planning of trajectories through third-order polynomials this resembles curves smoothed by Bezier to generate different deformations in the robot in such a way that its movements elude obstacles in a given one within the workspace. The developed algorithm was implemented on a simulated virtual platform in a robotics environment. The motivation of the work was to be able to demonstrate a planning of robot trajectories with multiple degrees of freedom using deterministic algorithms and not focused on computational intelligence such as neural networks or reinforced learning.

Seismic inversion is used in practice as a tool for predicting reservoir properties. It allows one to extract a model with a high level of detail from seismic data, i.e. high-frequency component of the model. In this case, the input data are the time processing results, and the issues related to the low-frequency component of the model are not considered usually. This work describes the implementation of a model-based seismic inversion algorithm. The input data for the inversion are the depth image results in true amplitudes and the depth migration velocity model. The possibilities of seismic inversion are numerically investigated to refine the low-frequency component of the model. Experiments were carried out using synthetic seismic data got for realistic Sigsbee model.

The Smoothed Particle Hydrodynamics (SPH) method is a meshless Lagrangian method widely used in continuum mechanics simulation. Despite its wide application, theoretical issues of SPH approximation, stability, and convergence are among the unsolved problems of computational mathematics. In this paper, we present the application of dispersion analysis to the SPH approximation of one-dimensional gas dynamics equations to study numerical phenomena that appeared in practice. We confirmed that SPH converges only if the number of particles per wavelength increases while smoothing length decreases. At the same time, reduction of the smoothing length when keeping the number of particles in the kernel fixed (typical convergence results for finite differences and finite elements) does not guarantee the convergence of the numerical solution to the analytical one. We indicate the particular regimes with pronounced irreducible numerical dispersion. For coarse resolution, our theoretical findings are confirmed in simulations.

Accuracy of measuring and observation processes depends greatly on the stabilization of the appropriate equipment located on moving vehicles. We propose to design stabilization systems based on robust control that can ensure the required accuracy in difficult conditions of real operation. The main issue of the research is the development of numerical algorithms for designing robust stabilization systems assigned for control of inertial platforms motion. The analysis of applications and classification of inertially stabilized platforms is given. The block diagram of the algorithm of the robust parametrical optimization is represented. Features of this numerical algorithm are discussed including forming the optimization criterion and implementation of the optimization procedure. The block diagram of the robust structural synthesis is represented. Features of forming the function of mixed sensitivity are given. Results of simulation for the inertially stabilized platforms assigned for the operation of the ground moving vehicles are shown.

We introduce a new class of simultaneously diagonalizable real matrices, the $$\gamma $$ γ -matrices, which include both symmetric circulant matrices and a subclass of the set of all reverse circulant matrices. We define some algorithms for fast computation of the product between a $$\gamma $$ γ -matrix and a real vector. We proved that the computational cost of a multiplication between a $$\gamma $$ γ -matrix and a real vector is of at most $$\frac{7}{4} \, n \, \log _2 n+o( n \, \log _2 n)$$ 7 4 n log 2 n + o ( n log 2 n ) additions and $$\frac{1}{2} \, n \, \log _2 n+o( n \, \log _2 n)$$ 1 2 n log 2 n + o ( n log 2 n ) multiplications. Our algorithm can be used to improve the performance of general discrete transforms for multiplications of real vectors.

The reliability and transmission quality in any wireless communication network is paramount. Unmanned Aerial Vehicle (UAV) communication networks are no exception to this fact. Hence, the need to investigate various communication technologies that will improve and be able to support the various applications of the UAV communication network. Correlated based stochastic models (CBSM) have been used to assess theoretical performances in UAV communication networks. CBSM has insufficient precision in a practical system. Geometry based stochastic channel models (GBSM) on the other hand, displays realistic channel features. These realistic channel features include pathloss, angle of arrival (AoA), angle of departure (AoD), etc. GBSM is better and ideal for channel modeling. This paper analyses the UAV communication networks in terms of their reliability and quality in transmission. MIMO-OFDM technology has been proposed to improve the UAV communication network. In this UAV network, the transmitters are modeled as cylindrical array (CA) because of its attribute of good regulation in 3-D space among others. A 3-D GBSM is proposed in the analysis. Also, an interference model has been presented in the UAV communication network. Results from this research show that MIMO-OFDM improves the reliability and quality of the UAV communication network. Generally, the capacity and BER increased with increasing number of antennas and SINR. However, beyond SINR of 25dB, we observed an irreducible error floor, that is, BER remained constant with increasing SINR. Successive Interference Cancellation (SIC) was therefore employed to minimize the irreducible error floor in the UAV communication network. This increased the average capacity and BER to about 25bits/s/Hz and $$10^{-8}$$ 10 - 8 respectively.

Cloud computing has become a dominating trend in the IT industry and academia. In recent years, many Cloud providers have emerged in the market, each one has proper standards and service interface. Migration of legacy systems and their data to a Cloud platform or moving applications from one Cloud provider to another are complicated and high cost processes. In major legacy on-premise applications, they bring technical and business challenges. The choice of the best migration strategy is a hard task for Small and Medium-Sized Enterprises (SMEs) looking at best price and zero risk. Although guidelines offered by Cloud providers to support their users in migration tasks, the user implication is often mandatory to perform a conclusive migration. In this paper, we provide an overview of contemporary approaches for the model based migration of applications into Cloud computing. We highlight the Pros and Cons of each one and we present a summary comparison of these approaches. As well, we propose a new Cloud migration approach that renders the migration process more straightforward than with existent approaches and guarantees the services and data portability.

The expansion of networks worldwide and its benefits allowed the emergence and application of several technologies aimed at exchanging information in a decentralized manner, such as Blockchain, where there is no need for a reliable agent or third party for management. In this context, the emergence of various distributed wireless networks was possible, such as community networks, where large mesh networks composed of hundreds of nodes built by communities seek to solve digital inclusion problems. Wireless mesh networks still present some challenges in their growth. Even with specific protocols like Babel, which optimize its operation, there are problems such as reliability among network members and the correct distribution of resources to encourage its growth. These characteristics prevent wireless mesh networks from developing to reach areas of technology scarcity called the last mile. Thus, this work uses the optimization techniques for WMNs achieved by the Babel protocol and proposes the modification of the original Babel protocol to insert financial incentive mechanisms into the protocol. The article presents the proposed modifications and demonstrates, through simulation tests, the functioning and efficiency of the protocol extensions through the OMNeT++ platform combined with the INET Framework tool.

In a distributed mobile Software Defined Mobile Network (SDMN), a device is in a continuous movement from one domain to another. Since a different controller is responsible for each domain, the devices’ authentication information needs to be communicated between the SDN controllers (this type of communication is called East-West communication) to ensure the continuity access to the used devices services. This paper proposes a new secure East-West communication to approve service access continuity for mobile devices in a Distributed SDN. A comparative study between the VPN based approach and the proposed one is done both analytically and by simulation according to two criteria: the approval access authentication delay and the communication overhead. The results show that the proposed approach provides a secure East-West communication using less communication overhead and decreases the approval access authentication delay of mobile devices in an SDN.

The multi-input/multi-output (MIMO) channel 2D convolution is the most compute-intensive operation in Convolutional Neural Networks (CNNs). This paper presents a high-performance implementation for the MIMO convolution by extending the well-known software systolic array model (SSAM) in which the partially computed results are shifted or shuffled across multiple threads in a CUDA warp to compute the single-input/single-output (SISO) channel convolution. We propose two methods for computing a full MIMO convolution on the GPU system. In the first method, the MIMO convolution is performed by iterations of the multi-input/single-output (MISO) convolution across multiple output channels while the second method iterates the single-input/multi-output (SIMO) convolution across multiple input channels. Both methods systolically shuffle partial results multiple times during the MIMO computing. It is shown that the first method mostly demonstrates a higher performance than the second one, since the first one reuses data effectively on the L1/L2 caches as well as on the register files. We also experimentally demonstrate that a single-precision performance of the directly implemented MIMO convolution is much better than that of the SSAM/SISO-based convolution and a GEMM-based MIMO convolution of the NVIDIA cuDNN library.

Proposals for Information Centric Networking (ICN) have recently emerged to rethink the foundations of the Internet and design a native data-oriented network architecture. Among the current ICN projects, Named Data Networking (NDN) is a promising architecture supported by the National Science Foundation (NSF). The NDN communication model is based on the Publish/Subscribe paradigm and focuses on broadcasting and finding content and introduces caching in intermediate routers. Data packets are sent in response to a prior request called an Interest packet and the data are cached along the way to the original requester. Content caching is an essential component of NDN in order to reduce bandwidth consumption and improve data delivery speed, however, this feature allows malicious nodes to perform attacks that are relatively simple to implement but very effective. For that reason, the goal of this paper is to study and classify the types of attacks that can target the NDN architecture such as (Cache Pollution Attack (CPA), Cache Poisoning Attack, Cache Privacy Attack, Interest Flooding Attack (IFA), etc.) according to their consequences in terms of reducing the performance of the network. Moreover, we give an overview about the proposed detection mechanisms and their limitations.

In today’s era, more and more businesses are moving on the cloud. The customer needs to be assured of the quality-of-service they desire from the cloud service provider. Service level agreement (SLA) is a significant medium for building this confidence. The services are often running on virtual machines (VMs) provided by service provider. Various cloud management activities require transferring of VMs among the servers. They make use of live migration to achieve these objectives. With multiple VMs being migrated, it is essential to assign the available bandwidth among the VMs to minimize the penalty incurred due to SLA violation during migration. However, depending upon the assigned transfer rates, some VMs in the group might complete the migration process earlier than the others. In this paper, we propose an algorithm and a controller to assign freed bandwidth to the remaining VMs in the group. With extensive simulations, it is found that the penalty, migration time, and downtime for VMs is significantly reduced compared to the state-of-the-art approaches.

This paper applies the MPI-OpenMP-based two-dimensional Hopmoc method using the explicit work-sharing technique with a recently proposed mechanism to reduce implicit barriers in OpenMP. Specifically, this paper applies the numerical algorithm to yield approximate solutions to the advection-diffusion equation. Additionally, this article splits the mesh used by the numerical method and distributes them to over-allocated threads. The mesh partitions became so small that the approach reduced the cache miss rate. Consequently, the strategy accelerated the numerical method in multicore systems. This paper then evaluates the results of implementing the strategy under different metrics. As a result, the set of techniques improved the performance of the parallel numerical method.

During the end of 2017, the systems developed by policemen in the Intelligence Department of São Paulo State Military Police were outdated in technology and did not follow any kind of pattern, neither had integration with services, databases and other systems, being very difficult to maintain. However, this kind of system could not be abandoned, because they were important in the organization. Thus, the managers of the Intelligence Department understood the need for modernization of architecture and system technologies. There are some approaches that can be chosen in this evolution process. A difficult question to be answered is related to which path is better to be followed in this process of system migration, mainly about the choosing of monolithic or microservices architecture. Although the microservice architecture is modern, it is not so easy to be implemented directly in a legacy system evolution process, because it depends on a strong domain of businesses areas and technology, which makes it difficult to involve of beginners developers. In this paper, we reported an experience about migrating a legacy system used in Intelligence Department of Sao Paulo State Military Police to a microservice-based architecture by re-engineering the legacy system as a monolith first and then decomposing it to microservices. We realized that microservices did not necessarily had to be adopted in the first place, but it could be a natural consequence of evolution, using a monolithic approach in the beginning. This process may also help the development team to improve their knowledge about the adopting of new technologies, architectures and better understanding of business.

Eye centre localisation is critical to eye tracking systems of various forms and with applications in variety of disciplines. An active eye tracking approach can achieve a high accuracy by leveraging active illumination to gain an enhanced contrast of the pupil to its neighbourhood area. While this approach is commonly adopted by commercial eye trackers, a dependency on IR lights can drastically increase system complexity and cost, and can limit its range of tracking, while reducing system usability. This paper investigates into a passive eye centre localisation approach, based on a single camera, utilising convolutional neural networks. A number of model architectures were experimented with, including the Inception-v3, NASNet, MobileNetV2, and EfficientNetV2. An accuracy of 99.34% with a 0.05 normalised error was achieved on the BioID dataset, which outperformed four other state-of-the-art methods in comparison. A means to further improve this performance on high-resolution data was proposed; and it was validated on a high-resolution dataset containing 12,381 one-megapixel images. When assessed in a typical eye tracking scenario, an average eye tracking error of 0.87% was reported, comparable to that of a much more expensive commercial eye tracker.

Among the geometric bodies of revolution we find the torus of revolution generated from a circumference that rotates around an axis. Given the classic definition used in Mathematics, interest arises in finding other curves that generate the torus of revolution when rotating around an axis. There is already work done, about the construction of toruses of revolution, using a lemniscatic curve. In this article, making the respective analysis and the necessary programming using the Mathematica 11.1 software, allowed us to carry out the necessary calculations and geometric visualizations of the mathematical object: So a torus of revolution was built from the curve of eight in its parametric form and even the equation of the torus in its Cartesian form. The study was extended and the torus of revolution was generated from rational and irrational curves that rotate around an axis. Curves were determined that were on the torus generated by a curve of eight, which when properly projected to planes, curves that have symmetries were obtained. When points on these curves are properly taken, special irregular polygons are obtained. By obtaining these results, a satisfactory answer to the research question was obtained, as well as a way to define it. In addition, it has shown us a wide path of research on the different curves that can generate a torus of revolution.

Polygonal meshes are used in CAD/CAM systems and in solutions of many of engineering problems. Many of those rely on polygonal representation using facets, edges and vertices. Today, due to numerical robustness as only three points can lie on a plane, limited numerical precision of the floating point representation, etc. the triangular facets are used nearly exclusively. This is a significant factor witch is not fully considered in triangular mesh representations and their processing. This contribution presents a new approach to the 3D geometric model representation based on vertices and edges only, i.e. by the wire-frame data model, where no facet representation is needed, if the surface is formed by a triangular mesh. The wire-frame representation use leads to significant reduction of data as there is no need to represent facets explicitly. It can be used for significant data compression, etc. Examples demonstrating the worst cases solutions are presented with a 3D print of those.

The coronavirus outbreak became a major concern for society worldwide. Technological innovation and ingenuity are essential to fight COVID-19 pandemic and bring us one step closer to overcome it. Researchers over the world are working actively to find available alternatives in different fields, such as the Healthcare System, pharmaceutic, health prevention, among others. With the rise of artificial intelligence (AI) in the last 10 years, IA-based applications have become the prevalent solution in different areas because of its higher capability, being now adopted to help combat against COVID-19. This work provides a fast detection system of COVID-19 characteristics in X-Ray images based on deep learning (DL) techniques. This system is available as a free web deployed service for fast patient classification, alleviating the high demand for standards method for COVID-19 diagnosis. It is constituted of two deep learning models, one to differentiate between X-Ray and non-X-Ray images based on Mobile-Net architecture, and another one to identify chest X-Ray images with characteristics of COVID-19 based on the DenseNet architecture. For real-time inference, it is provided a pair of dedicated GPUs, which reduce the computational time. The whole system can filter out non-chest X-Ray images, and detect whether the X-Ray presents characteristics of COVID-19, highlighting the most sensitive regions.

We deal with the problem of blind separation of the components, in particular for documents corrupted by bleed-through and show-through. So, we analyze a regularization technique, which estimates the original sources, the interference levels and the blur operators. We treat the estimate of the interference levels, given the original sources and the blur operators. In particular, we investigate several GNC-type algorithms for minimizing the energy function. In the experimental results, we find which algorithm gives more precise estimates of the interference levels.

In this paper we study a Blind Source Separation (BSS) problem, and in particular we deal with document restoration. We consider the classical linear model. To this aim, we analyze the derivatives of the images instead of the intensity levels. Thus, we establish non-overlapping constraints on document sources. Moreover, we impose that the rows of the mixture matrices of the sources have sum equal to 1, in order to keep equal the lightnesses of the estimated sources and those of the data. Here we give a technique which uses the symmetric factorization, whose goodness is tested by the experimental results.

Regardless of the collection location, mobile traffic data contains information about many aspects of subscribers’ lives, including their activities, interests, schedules, travel and preferences. It is precisely the ability to access such information on unprecedented scales that is of critical importance for studies in a wide variety of fields. However, access to such a rich source also raises concerns about potential infringements on the rights of mobile customers regarding their personal data: among other things, individuals can be identified, their movements can be modified, their movements can be tracked and their mobile stage fright can be monitored. As a result, regulators have been working on legislation to protect the privacy of mobile users. In this optic, we provide a scalable solution to anonymize Big Data Spatio-temporal Trajectories of mobile users.

Big Data usually refers to the unprecedented growth of data and associated processes to gather, store, process, and analyze them to provide organizations and users with useful insights and information. The intrinsic complexity and characteristics of systems handling Big Data require software architectures as founded drivers for these systems to meet functional and quality requirements. In light of the relevant role of software architectures for Big Data systems, we investigate the current state of the art of Big Data software architectures. This paper presents the results of a systematic mapping study that updates existing literature reviews on this topic. We selected and analyzed 23 primary studies published in the last five years. We identified 11 architecture-related quality requirements and six architectural modules relevant to the design of software architectures for Big Data systems, besides analyzing whether existing proposals of reference architectures comply with these requirements and modules. We expect the results presented in this paper can provide a continuous update of the state of the art while highlighting essential concerns in the design of software architectures for Big Data systems.

Speech recognition is a technology that aid processing of speech signals through communicating with computer applications. Previous studies exhibits speech recognition errors arising from users’ acoustic irrational behavior. This research paper provides acoustic nudging-based model for reformulating the persistence of automatic speech recognition errors that involve the user’s acoustic irrational behavior and distortion of speech recognition accuracy. Gaussian mixture model (GMM) helped in addressing the low-resourced attribute of Yorùbá language to achieve better accuracy and system performance. From the implemented results, it was observed that the proposed acoustic nudging-based model improves accuracy and system performance based on Word Error Rate (WER), validation, testing and training accuracy. The evaluation results for the mean WER was 4.723% when compared to existing models. This approach thereby reduces error rate when compared with previous models by GMM (1.1%), GMM-HMM (0.5%), CNN (0.8%), and DNN (1.4%). Therefore, this work was able to discover a foundation for advancing the current understanding of under-resourced languages and development of an accurate and precise model for speech recognition.

Lack of methodologies for ontology quality evaluation causes a challenge in producing good quality ontologies. Thus, we developed an iterative quality methodology to address this gap by analyzing the existing quality theories defined in ontology engineering, as well as, the theories in software engineering. Accordingly, this paper presents the developed methodology including how the other ontology quality theories get associated with it. Moreover, a use case in the agriculture domain has been demonstrated in order to provide an understanding of how the methodology can be applied in a real context. In the future, many experiments are expected to be carried out to fine-tune the methodology and to illustrate its usefulness.

Nowadays Indoor Positioning Systems (IPS) are attracting attention in literature because of Global Positioning System (GPS) challenge to track and navigate indoors. These IPSs intend to provide information about a wireless object’s current position indoor. GPS-based localization is the most successful Location-Based Service (LBS) application deployed in an outdoor environment. However, GPS faces a challenge of the line of sight indoor. GPS is affected extensively by the multipath-effects. IPS technologies such as Wi-Fi are deployed for indoor localization, in an attempt to alleviate the GPS indoor challenges. Most IPS employs the Fingerprinting algorithm, whereby a radio-map is generated during the offline phase by collecting measurements of Received Signal Strength Indicator (RSSI) at known locations and, positioning of devices at an unknown location is performed during the online phase by utilizing Machine Learning (ML) Algorithms. The radio-map dataset plays a major role in the accuracy performance of the classifiers deployed during the online phase. RSSI fluctuates indoors because of fading, interferences, and shadowing, therefore, the correction of the radio-map RSSI measurements is mandatory to improve the classifiers performance. In this paper, we looked into the impact of the size of the calibration radio-map on the accuracy of the predictive model. We applied the Mean and Standard Deviation filter on three datasets of different sizes to reduce the RSSI instability at each required point and conducted comparative performance on how ML classification models perform on the three radio-map different in size. The radio-map was generated using our EMPsys application. The results of the simulations show that the accuracy of the Kernel Naïve Bayes significantly improved with filter as the radio-map size increased, from 64.4% with 453 observations in the first scenario to 95.4% with 1054 observations in the third scenario. We, therefore, conclude that the performance of the classifier to be used during the online phase of the fingerprinting algorithm relies on both the size of the radio-map and the filtering methods used to correct the RSSI measurements of the radio-map.

Breast cancer is the most publicized cancer that hits women around the world. It’s considered as the second cause of death among females. Early detection helps a lot in increasing the survival rate, and the probability of recovery from this disease. The mammogram is the main screening modality that is used regularly for breast cancer diagnosis. The accurate interpretation of the mammogram is very important for mass detection and diagnosis. The rapid evolution of deep learning is contributing to introduce more accurate systems that can act as a second opinion for the radiologists, and accordingly, this can help in providing an accurate diagnosis. In this paper, we propose a model for mass detection and classification based on You Look Only Once (YOLO)v4. We designed the experiment to investigate the performance of different augmentation techniques using YOLOv4 including mosaic that was introduced by YOLOv4. Furthermore, in the preprocessing phase, the images were reconstructed to be in a multichannel format which enhanced the detection accuracy by almost $$\simeq$$ ≃ 10%. The model was evaluated with the usage of different combinations of augmentation techniques (mosaic, mix-up, and conventional augmentation). The experiments were conducted on the INbreast and MIAS datasets, the results of INbreast showed that mosaic with YOLOv4 achieved the best results with mAP (mean average precision), precession, and recall of almost $$\simeq$$ ≃ 99.5%, 98%, and 94% respectively for detection. In addition, the proposed model achieved AP of 99.16% and 99.58% for classifying the detected masses as benign and malignant respectively. Furthermore, the best results on MIAS achieved mAP, precession, and recall of 95.28%, 93%, and 90% respectively. Finally, our methodology showed competitive performance compared to other similar studies.

The recent success of distinct e-commerce systems has driven many fashion companies into the online marketplace, allowing consumers to quickly and easily access a worldwide network of brands. However, to succeed in this scenario, it is necessary to provide a tailored, personalized, and reliable fashion shopping experience. Moreover, unfortunately, current solutions on marketing have provided a general approach to push and suggest the most popular or purchased items in most cases. Thus, this paper proposes a new ensemble recommendation system based on the stacking of classical approaches associated with contextual information about customers and products. Our idea is to incorporate user preferences and item characteristics to ensure a desirable level of personalization to commonly applied methods. Our method is a Neural Collaborative Filtering algorithm that can combine any recommendation system with contextual domain information. The results are promising, showing significant gains of up to 80% MRR, 70% NDCG, and 108% Hits when compared to popular baselines for this same scenario.

The paper describes the origin and evolution of the monitoring system of the Duomo di Milano, which was installed during the 1960s. In that period, differential movements induced by the extraction of groundwater (among other factors) caused significant instability and risk of collapse of the monument. Today, the monitoring system is still operative. Instruments, techniques, and calculation procedures were continuously updated considering the continuity of the time series, resulting in a precious archive for structural health monitoring and conservation. The actual configuration of the monitoring system includes a large variety of both automatic and manual sensors, digital and mechanical, in real-time or with an established periodicity. Moreover, the monitoring system inside the cathedral still preserves continuity with the original measurements thanks to continuous maintenance carried out over time. The manuscript illustrates and discusses the original system as well as the updates related to activities that began more than half a century ago.

Markov Decision Process (MDP) is the most common planning framework in the literature for sequential decisions under probabilistic outcomes; MDPs also underlies the Reinforcement Learning (RL) theory. Computerized Adaptive Testing (CAT) is an assessment approach that selects questions one after another while conditioning each selection on the previous questions and answers. While MDP defines a well-posed optimization planning-problem, shortsighted score functions have solved the planning problem in CATs. Here, we show how MDP can model different formalisms for CAT, and, therefore, why the CAT community may benefit from MDP algorithms and theory. We also apply an MDP algorithm to solve a CAT, and we compare it against traditional score functions from CAT literature.

Sign languages are visual representations used by hearing or speech impaired people to communicate between themselves and with other people. There are over 140 sign languages globally, and they can be developed by deaf communities or derived from other existing sign languages. The signs made in this context are not considered gestures but words articulated primarily by the hands, while possibly involving facial expressions and trunk movements, making it far from trivial to understand them. Thus, automatic sign language recognition supported by machine learning has shown significant advancement in recent years. This paper presents a mobile application capable of recognizing gestures representing the letters of the Brazilian Sign Language’s (Libras) alphabet. Our methodology has three steps: the construction of the gesture dataset containing all the letters of the Libras alphabet; the training of the machine learning model used in the gesture classification; and the development of the desktop/mobile application used to capture the gestures to be classified.

Recent advances in machine learning and computer vision have led to widespread use of these technologies in the industrial sector. Quality control and production counting are the most important applications. This article describes a solution for counting products in an industrial production line. It consists of two main modules: i) hardware infrastructure and ii) software solution. In ii) there are modules for image capture and product recognition using the Yolov5 algorithm and modules for tracking and counting products. The results show that our solution achieves $$99.91\%$$ 99.91 % accuracy in product counting and classification. Furthermore, these results were compared to the current manual counting system used in the industry considered in this study. This demonstrated the feasibility of our solution in a real production environment.

This paper presents a voice gender recognition system. Acoustic features and Mel-Frequency Cepstral Coefficients (MFCCs) are extracted to define the speaker's gender. The most used features in these kinds of studies are acoustic features, but in this work, we combined them with MFCCs to test if we will get more satisfactory results. To examine the performance of the proposed system we tried four different databases: the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), the Saarbruecken Voice Database (SVD), the CMU_ARCTIC database and the Amazigh speech database (Self-Created). At the pre-processing stage, we removed the silence from the signals by using Zero-Crossing Rate (ZCR), but we kept the noises. Support Vector Machine (SVM) is used as the classification model. The combination of acoustic features and MFCCs achieves an average accuracy of 90.61% with the RAVDESS database, 92.73% with the SVD database, 99.87% with the CMU_ARCTIC database and 99.95% with the Amazigh speech database.

Our previous study on classification and detection of abusive language in South African social media space has shown the high prospect of surface level features and classical machine learning models in terms of accuracy. However, much improvement is still needed in the aspect of F1-score. Therefore, the state-of-the-arts such as neural embeddings (Word2Vec, Doc2Vec, GloVe) and neural network-based transformer models (BERT and mBERT) which have performed well in many hate speech isolation tasks are explored in this work. In the evaluation of classical machine learning algorithms, Word2Vec with Support Vector Machine outperformed the previous models, Doc2Vec and GloVe in terms of F1-score. In the evaluation of neural networks, all the neural embedding and transformer models performed worse than the previous models in terms of F1-score. In conclusion, the impressive performance of Word2Vec neural embedding with classical machine learning algorithms in terms of best F1-score of 0.62 and accuracy of 0.86 shows its good prospect in the isolation of abusive language and hate speech in South African social media space.

Open source communities as large-scale paradigm innovations have attracted a lot of attention from scholars. In this paper, we analyze the possible effects that software companies' participation in open source communities can have on corporate patents through knowledge flow theory. Technological innovation resulting from the acquisition of technical knowledge by open source communities shortens the cycle of corporate technological innovation. However, since firms' technological innovation activities can only be imitations and follow-the-flow innovations, they may sometimes cause the failure of firms' technological innovation.