Skip to main content

Über dieses Buch

This book proposes neural networks algorithms and advanced machine learning techniques for processing nonlinear dynamic signals such as audio, speech, financial signals, feedback loops, waveform generation, filtering, equalization, signals from arrays of sensors, and perturbations in the automatic control of industrial production processes. It also discusses the drastic changes in financial, economic, and work processes that are currently being experienced by the computational and engineering sciences community.

Addresses key aspects, such as the integration of neural algorithms and procedures for the recognition, the analysis and detection of dynamic complex structures and the implementation of systems for discovering patterns in data, the book highlights the commonalities between computational intelligence (CI) and information and communications technologies (ICT) to promote transversal skills and sophisticated processing techniques.

This book is a valuable resource for

a. The academic research community

b. The ICT market

c. PhD students and early stage researchers

d. Companies, research institutes

e. Representatives from industry and standardization bodies





Chapter 1. Processing Nonlinearities

The problem of non-linear data is one of the oldest in experimental science. The solution to this problem is very complex, since the exact mechanisms that describe a phenomenon and its nonlinearities, are often unknown. At the same time, environmental factors such as the finite precision of the processing machine, noise, and sensor limitations—among others—produce further inaccuracies making even more unfitting the description of the phenomenon described by the collected data. In this context, while developing complex systems, with optimal performance, capable of interacting with the environment in an autonomous way, and showing some form of intelligence, the ultimate solution is to process, identify and recognize such nonlinear dynamics. Problems and challenges in Computational Intelligence (CI) and Information Communication Technologies (ICT) are devoted to implement sophisticated detection, recognition, and signal processing methodologies, to promptly, efficiently and effectively manage such problems. To this aim, neural networks, deep learning networks, genetic algorithms, fuzzy logic, and complex artificial intelligence designs, are favored because of their easy handling of nonlinearities while discovering new data structure, and new original patterns to enhance the efficiency of industrial and economic applications. The collection of chapters presented in this book offer a scenery of the current progresses in such scientific domain.
Anna Esposito, Marcos Faundez-Zanuy, Francesco Carlo Morabito, Eros Pasero

Processing Nonlinearities


Chapter 2. Temporal Artifacts from Edge Accumulation in Social Interaction Networks

There has been extensive research on social networks and methods for specific tasks such as: community detection, link prediction, and tracing information cascades; and a recent emphasis on using temporal dynamics of social networks to improve method performance. The underlying models are based on structural properties of the network, some of which we believe to be artifacts introduced from common misrepresentations of social networks. Specifically, representing a social network or series of social networks as an accumulation of network snapshots is problematic. In this paper, we use datasets with timestamped interactions to demonstrate how cumulative graphs differ from activity-based graphs and may introduce temporal artifacts.
Matt Revelle, Carlotta Domeniconi, Aditya Johri

Chapter 3. Data Mining by Evolving Agents for Clusters Discovery and Metric Learning

In this paper we propose a novel evolutive agent-based clustering algorithm where agents act as individuals of an evolving population, each one performing a random walk on a different subset of patterns drawn from the entire dataset. Such agents are orchestrated by means of a customised genetic algorithm and are able to perform simultaneously clustering and feature selection. Conversely to standard clustering algorithms, each agent is in charge of discovering well-formed (compact and populated) clusters and, at the same time, a suitable subset of features corresponding to the subspace where such clusters lie, following a local metric learning approach, where each cluster is characterised by its own subset of relevant features. This not only might lead to a deeper knowledge of the dataset at hand, revealing clusters that are not evident when using the whole set of features, but can also be suitable for large datasets, as each agent processes a small subset of patterns. We show the effectiveness of our algorithm on synthetic datasets, remarking some interesting future work scenarios and extensions.
Alessio Martino, Mauro Giampieri, Massimiliano Luzi, Antonello Rizzi

Chapter 4. Neural Beamforming for Speech Enhancement: Preliminary Results

In the field of multi-channel speech quality enhancement, beamforming algorithms play a key role, being able to reduce noise and reverberation by spatial filtering. To that extent, an accurate knowledge of the Direction of Arrival (DOA) is crucial for the beamforming to be effective. This paper reports extremely improved DOA estimates with the use of a recently introduced neural DOA estimation technique, when compared to a reference algorithm such as Multiple Signal Classification (MUSIC). These findings motivated for the evaluation of beamforming with neural DOA estimation in the field of speech enhancement. By using the neural DOA estimation in conjunction with beamforming, speech signals affected by reverberation and noise improve their quality. These first findings are reported to be taken as a reference for further works related to beamforming for speech enhancement.
Stefano Tomassetti, Leonardo Gabrielli, Emanuele Principi, Daniele Ferretti, Stefano Squartini

Chapter 5. Error Resilient Neural Networks on Low-Dimensional Manifolds

We introduce an algorithm that improves Neural Network classification/registration of corrupted data belonging to low-dimensional manifolds. The algorithm combines ideas of the Orthogonal Greedy Algorithm with the standard gradient back-propagation engine incorporated in Neural Networks. Therefore, we call it the Greedient algorithm.
Alexander Petukhov, Inna Kozlov

Chapter 6. FIS Synthesis by Clustering for Microgrid Energy Management Systems

Microgrids (MGs) are the most affordable solution for the development of smart grid infrastructures. They are conceived to intelligently integrate the generation from Distributed Energy Resources (DERs), to improve Demand Response (DR) services, to reduce pollutant emissions and curtail power losses, assuring the continuity of services to the loads as well. In this work it is proposed a novel Fuzzy Inference System (FIS) synthesis procedure as the core inference engine of an Energy Management System (EMS) for a grid-connected MG equipped with a photovoltaic power plant, an aggregated load and an Energy Storage System (ESS). The EMS is designed to operate in real time by defining the ESS energy flow in order to maximize the revenues generated by the energy trade with the distribution grid considering a Time Of Use (TOU) energy prices policy. The FIS adopted is a first order Tagaki-Sugeno type, designed through a data driven approach. In particular, multidimensional Membership Functions (MFs) are modelled by a K-Means clustering algorithm. Successively, each cluster is used to define both the antecedent and the consequent parts of a tailored fuzzy rule, by estimating a multivariate Gaussian MF and the related interpolating hyperplane. Results have been compared with benchmark references obtained by a Linear Programming (LP) optimization. The best solution found is characterized by a small number of MFs, namely a limited number of fuzzy rules. Its performances are close to the optimum solution in terms of profit generated and, moreover, it shows a smooth exploitation of the ESS.
Stefano Leonori, Maurizio Paschero, Antonello Rizzi, Fabio Massimo Frattale Mascioli

Chapter 7. Learning Activation Functions from Data Using Cubic Spline Interpolation

Neural networks require a careful design in order to perform properly on a given task. In particular, selecting a good activation function (possibly in a data-dependent fashion) is a crucial step, which remains an open problem in the research community. Despite a large amount of investigations, most current implementations simply select one fixed function from a small set of candidates, which is not adapted during training, and is shared among all neurons throughout the different layers. However, neither two of these assumptions can be supposed optimal in practice. In this paper, we present a principled way to have data-dependent adaptation of the activation functions, which is performed independently for each neuron. This is achieved by leveraging over past and present advances on cubic spline interpolation, allowing for local adaptation of the functions around their regions of use. The resulting algorithm is relatively cheap to implement, and overfitting is counterbalanced by the inclusion of a novel damping criterion, which penalizes unwanted oscillations from a predefined shape. Preliminary experimental results validate the proposal.
Simone Scardapane, Michele Scarpiniti, Danilo Comminiello, Aurelio Uncini

Chapter 8. Context Analysis Using a Bayesian Normal Graph

Contextual information can be used to help object detection in video and images, or to categorize text. In this work we demonstrate how the Latent Variable Model, expressed as a Factor Graph in Reduced Normal Form, can manage contextual information to support a scene understanding task. In an unsupervised scenario our model learns how various objects can coexist, by associating object variables to a latent Bayesian cluster. The model, that is implemented using probabilistic message propagation, can be used to correct or to assign labels to new images.
Amedeo Buonanno, Paola Iadicicco, Giovanni Di Gennaro, Francesco A. N. Palmieri

Chapter 9. A Classification Approach to Modeling Financial Time Series

In this paper, several classification methods are applied for modeling financial time series with the aim to predict the trend of successive prices. By using a suitable embedding technique, a pattern of past prices is assigned a class if the variation of the next price is over, under or stable with respect to a given threshold. Furthermore, a sensitivity analysis is performed in order to verify if the value of such a threshold influences the prediction accuracy. The experimental results on the case study of WTI crude oil commodity show a good classification accuracy of the next (predicted) trend, and the best performance is achieved by the K-Nearest Neighbors classification strategy.
Rosa Altilio, Giorgio Andreasi, Massimo Panella

Chapter 10. A Low-Complexity Linear-in-the-Parameters Nonlinear Filter for Distorted Speech Signals

In this paper, the problem of the online modeling of nonlinear speech signals is addressed. In particular, the goal of this work is to provide a nonlinear model yielding the best tradeoff between performance results and required computational resources. Functional link adaptive filters were proved to be an effective model for this problem, providing the best performance when trigonometric expansion is used as a nonlinear transformation. Here, a different functional expansion is adopted based on the Chebyshev polynomials in order to reduce the overall computational complexity of the model, while achieving good results in terms of perceived quality of processed speech. The proposed model is assessed in the presence of nonlinearities for both simulated and real speech signals.
Danilo Comminiello, Michele Scarpiniti, Simone Scardapane, Raffaele Parisi, Aurelio Uncini

Chapter 11. Hierarchical Temporal Representation in Linear Reservoir Computing

Recently, studies on deep Reservoir Computing (RC) highlighted the role of layering in deep recurrent neural networks (RNNs). In this paper, the use of linear recurrent units allows us to bring more evidence on the intrinsic hierarchical temporal representation in deep RNNs through frequency analysis applied to the state signals. The potentiality of our approach is assessed on the class of Multiple Superimposed Oscillator tasks. Furthermore, our investigation provides useful insights to open a discussion on the main aspects that characterize the deep learning framework in the temporal domain.
Claudio Gallicchio, Alessio Micheli, Luca Pedrelli

Chapter 12. On 4-Dimensional Hypercomplex Algebras in Adaptive Signal Processing

The degree of diffusion of hypercomplex algebras in adaptive and non-adaptive filtering research topics is growing faster and faster. The debate today concerns the usefulness and the benefits of representing multidimensional systems by means of these complicated mathematical structures and the criterions of choice between one algebra or another. This paper proposes a simple comparison between two isodimensional algebras (quaternions and tessarines) and shows by simulations how different choices may determine the system performance. Some general information about both algebras is also supplied.
Francesca Ortolani, Danilo Comminiello, Michele Scarpiniti, Aurelio Uncini

Chapter 13. Separation of Drum and Bass from Monaural Tracks

In this paper, we propose a deep recurrent neural network (DRNN), based on the Long Short-Term Memory (LSTM) unit, for the separation of drum and bass sources from a monaural audio track. In particular, a single DRNN with a total of six hidden layers (three feedforward and three recurrent) is used for each original source to be separated. In this work, we limit our attention to the case of only two, challenging sources: drum and bass. Some experimental results show the effectiveness of the proposed approach with respect to another state-of-the-art method. Results are expressed in terms of well-known metrics in the field of source separation.
Michele Scarpiniti, Simone Scardapane, Danilo Comminiello, Raffaele Parisi, Aurelio Uncini

Chapter 14. Intelligent Quality Assessment of Geometrical Features for 3D Face Recognition

This paper proposes a methodology to assess the discriminative capabilities of geometrical descriptors referring to the public Bosphorus 3D facial database as testing dataset. The investigated descriptors include histogram versions of Shape Index and Curvedness, Euclidean and geodesic distances between facial soft-tissue landmarks. The discriminability of these features is evaluated through the analysis of single block of features and their meanings with different techniques. Multilayer perceptron neural network methodology is adopted to evaluate the relevance of the features, examined in different test combinations. Principle Component Analysis (PCA) is applied for dimensionality reduction.
G. Cirrincione, F. Marcolin, S. Spada, E. Vezzetti

Computational Intelligence and Related Techniques in Industrial and ICT Engineering


Chapter 15. Convolutional Neural Networks for the Identification of Filaments from Fast Visual Imaging Cameras in Tokamak Reactors

The paper proposes a region-based deep learning convolutional neural network to detect objects within images able to identify the filamentary plasma structures that arise in the boundary region of the plasma in toroidal nuclear fusion reactors. The images required to train and test the neural model have been synthetically generated from statistical distributions, which reproduce the statistical properties in terms of position and intensity of experimental filaments. The recently proposed Faster Region-based Convolutional Network algorithm has been customized to the problem of identifying the filaments both in location and size with the associated score. The results demonstrate the suitability of the deep learning approach for the filaments detection.
Barbara Cannas, Sara Carcangiu, Alessandra Fanni, Ivan Lupelli, Fulvio Militello, Augusto Montisci, Fabio Pisano, Giuliana Sias, Nick Walkden

Chapter 16. Applying Network Analysis for Extracting Knowledge About Environment Changes from Heterogeneous Sensor Data Streams

Sensor network analysis has become a challenging task. The detection of sensor anomalies is one of the most prominent topics in this research area. In the past, researchers mainly focused on the detection and analysis of single-sensor anomalies. In this paper, we shift the focus from a local approach, aimed to detect anomalies on single sensors, to a global one, aiming at detecting and investigating the consequences, on the whole sensor network and/or its subnetworks, of anomalies present in one or more (heterogeneous) sensors.
Francesco Cauteruccio, Paolo Lo Giudice, Giorgio Terracina, Domenico Ursino

Chapter 17. Advanced Computational Intelligence Techniques to Detect and Reconstruct Damages in Reinforced Concrete for Industrial Applications

In reinforced concrete, as known, the steel bar, damping totally the traction stress, they are mainly subject to breakage. Then, as required by current legislation, it is necessary a check protocol of the specimens characterizing any defects since the typology of defect, often, determines its intended use. From this, the choice to use non-invasive technique such as Non-Destructive Testing and Evaluation (NDT / NDE) based on Eddy Currents is necessary. Starting from a campaign of Eddy Currents measurements potentially affected by uncertainty and/or imprecision, in this work we propose a new fuzzy approach based on Computing with Words techniques where a word is considered a label of a fuzzy set of points shared by similarities coming to an adaptive bank of fuzzy rules structured by classes possibly updated by the Expert’s knowledge. The numerical results obtained by means of the proposed approach are comparable with the results carried out by Fuzzy Similarities techniques already established in the literature.
Salvatore Calcagno, Fabio La Foresta

Chapter 18. Appraisal of Enhanced Surrogate Models for Substrate Integrate Waveguide Devices Characterization

Nowadays the use of surrogate models (SMs) is becoming a common practice to accelerate the optimization phase of the design of microwave and millimeter wave devices. In order to further enhance the performances of the optimization process, the accuracy of the response provided by a SM can be improved employing a suitable output correction block, obtaining in this way a so-called enhanced surrogate model (ESM). In this paper a comparative study of three different techniques for building ESMs, i.e. Kriging, Support Vector Regression Machines (SVRMs) and Artificial Neural Networks (ANNs), applied to the modelling of substrate integrated waveguide (SIW) devices, is presented and discussed.
Domenico De Carlo, Annalisa Sgrò, Salvatore Calcagno

Chapter 19. Improving the Stability of Variable Selection for Industrial Datasets

Variable reduction is an essential step in data mining, which is able effectively to increase both the performance of machine learning and the process knowledge by removing the redundant and irrelevant input variables. The paper presents a variable selection approach merging the dominating set procedure for redundancy analysis and a wrapper approach in order to achieve an informative and not redundant subset of variables improving both the stability and the computational complexity. The proposed approach is tested on different datasets coming from the UCI repository and from industrial contexts and is compared to the exhaustive variable selection approach, which is often considered optimal in terms of system performance. Moreover the novel method is applied to both classification and regression procedures.
Silvia Cateni, Valentina Colla, Vincenzo Iannino

Chapter 20. Cause and Effect Analysis in a Real Industrial Context: Study of a Particular Application Devoted to Quality Improvement

This paper presents an analysis of the occurrence of ripple defects during Hot Deep Galvanising of flat steel products, with a focus on the study on thick coils having low zinc coating. Although skilled personnel can manage ripples defects through particular operations, for instance wiping nitrogen instead of air in air blades, the real effects of each process parameter variation is unknown. Therefore, the study of these phenomena can improve the quality of coils, by decreasing reworked or scrapped material and reducing costs related to a redundant use of nitrogen. An accurate pre-processing procedure has been performed and then the analysis focused on the possible causes of ripples occurrences. In particular, the attention is focused on the development of a model capable to identify process variables with a stronger impact on the presence or absence of ripples, by expressing such effect through an appropriate relationship.
Silvia Cateni, Valentina Colla, Antonella Vignali, Jens Brandenburger

Chapter 21. An Improved PSO for Flexible Parameters Identification of Lithium Cells Equivalent Circuit Models

Nowadays, the equivalent circuit approach is one of the most used methods for modeling electrochemical cells. The main advantage consists in the beneficial trade-off between accuracy and complexity that makes these models very suitable for the State of Charge (SoC) estimation task. However, parameters identification could be difficult to perform, requiring very long and specific tests upon the cell. Thus, a more flexible identification procedure based on an improved Particle Swarm Optimization that does not require specific and time consuming measurements is proposed and validated. The results show that the proposed method achieves a robust parameters identification, resulting in very accurate performances both in the model accuracy and in the SoC estimation task.
Massimiliano Luzi, Maurizio Paschero, Antonello Rizzi, Fabio Massimo Frattale Mascioli

Intelligent Tools for Decision Making in Economics and Finance


Chapter 22. Yield Curve Estimation Under Extreme Conditions: Do RBF Networks Perform Better?

In this paper we test the capability of Radial Basis Function (RBF) networks to fit the yield curve under extreme conditions, namely in case of either negative spot interest rates, or high volatility. In particular, we compare the performances of conventional parametric models (Nelson–Siegel, Svensson and de Rezende–Ferreira) to those of RBF networks to fit term structure curves. To such aim, we consider the Euro Swap–EUR003M Euribor, and the USDollar Swap (USD003M) curves, on two different release dates: on December 30th 2004 and 2016, respectively, i.e. under very different market situations, and we examined the various ability of the above–cited methods in fitting them. Our results show that while in general conventional methods fail in adapting to anomalies, such as negative interest rates or big humps, RBF nets provide excellent statistical performances, thus confirming to be a very flexible tool adapting to every market’s condition.
Alessia Cafferata, Pier Giuseppe Giribone, Marco Neffelli, Marina Resta

Chapter 23. New Challenges in Pension Industry: Proposals of Personal Pension Products

Within the current post-crisis economic environment, characterized by low growth and low interest rates, retirement and long-term saving represent a crucial challenge. Furthermore, the expansion of life expectancies modifies the demand of pension products and insurers and pension providers have to guarantee the sustainability and competitiveness of their products, in spite of the economic stagnation. Within the context of the personal pension products, in the paper we propose a new contract with profit participation, which consists in a deferred life annuity with variable benefits changing according with two dynamic financial elements: the periodic financial result of the invested fund year by year and the first order financial technical base checked at the beginning of predefined intervals all along the contract life. A numerical implementation explains the forecasted trend of the inflows and outflows connected to the contract under financial and demographic stochastic assumptions.
Valeria D’Amato, Emilia Di Lorenzo, Marilena Sibillo

Chapter 24. A PSO-Based Framework for Nonsmooth Portfolio Selection Problems

We propose a Particle Swarm Optimization (PSO) based scheme for the solution of a mixed-integer nonsmooth portfolio selection problem. To this end, we first reformulate the portfolio selection problem as an unconstrained optimization problem by adopting an exact penalty method. Then, we use PSO to manage both the optimization of the objective function and the minimization of all the constraints violations. In this context we introduce and test a novel approach that adaptively updates the penalty parameters. Also, we introduce a technique for the refinement of the solutions provided by the PSO to cope with the mixed-integer framework.
Marco Corazza, Giacomo di Tollo, Giovanni Fasano, Raffaele Pesenti

Chapter 25. Can PSO Improve TA-Based Trading Systems?

In this paper, we propose and apply a methodology to improve the performances of trading systems based on Technical Indicators. As far as the methodology is concerned, we take into account a simple trading system and optimize its parameters—namely, the various time window lengths—by the metaheuristic known as Particle Swarm Optimization. The use of a metaheuristic is justified by the fact that the involved optimization problem is complex (it is nonlinear, nondifferentiable and integer). Therefore, the use of exact solution methods could be extremely time-consuming for practical purposes. As regards the applications, we consider the daily closing prices of eight important stocks of the Italian stock market from January 2, 2001, to April 28, 2017. Generally, the performances achieved by trading systems with optimized parameters values are better than those with standard settings. This indicates that parameter optimization can play an important role.
Marco Corazza, Francesca Parpinel, Claudio Pizzi

Chapter 26. Asymmetry Degree as a Tool for Comparing Interestingness Measures in Decision Making: The Case of Bayesian Confirmation Measures

Bayesian Confirmation Measures are used to assess the degree to which an evidence E supports or contradicts a conclusion H, making use of prior probability P(H), posterior probability P(H|E) and of probability of evidence P(E). Many confirmation measures have been defined till now, their use being motivated in different ways depending on the framework. Comparisons of those measures have already been made but there is an increasing interest for a deeper investigation of relationships, differences and properties. Here we focus on symmetry properties of confirmation measures which are partly inspired by classical geometric symmetries. Measures which do not satisfy a specific symmetry condition may present a different level of asymmetry: we define an asymmetry measure, some examples of its evaluation providing a practical way to appraise the asymmetry degree for Bayesian Confirmation Measures that allows to uncover some of their features, similarities and differences.
Emilio Celotto, Andrea Ellero, Paola Ferretti

Chapter 27. A Method Based on OWA Operator for Scientific Research Evaluation

This paper proposes a model for faculty evaluation based on OWA aggregation operators. Our method permits to consider interactions among the criteria in a formal way, and, at the same time, to realize an easy approach to understand, implement and apply.
Marta Cardin, Giuseppe De Nadai, Silvio Giove

Chapter 28. A Cluster Analysis Approach for Rule Base Reduction

In this paper we propose an iterative algorithm for fuzzy rule base simplification based on cluster analysis. The proposed approach uses a dissimilarity measure that allows to assign different importance to values and ambiguities of fuzzy terms in antecedent and consequent parts of fuzzy rules.
Luca Anzilli, Silvio Giove
Weitere Informationen

Premium Partner