Skip to main content
Top

2021 | Book

Artificial Intelligence and Soft Computing

20th International Conference, ICAISC 2021, Virtual Event, June 21–23, 2021, Proceedings, Part I

Editors: Prof. Leszek Rutkowski, Dr. Rafał Scherer, Marcin Korytkowski, Witold Pedrycz, Prof. Ryszard Tadeusiewicz, Dr. Jacek M. Zurada

Publisher: Springer International Publishing

Book Series : Lecture Notes in Computer Science

insite
SEARCH

About this book

The two-volume set LNAI 12854 and 12855 constitutes the refereed proceedings of the 20th International Conference on Artificial Intelligence and Soft Computing, ICAISC 2021, held in Zakopane, Poland, in June 2021. Due to COVID 19, the conference was held virtually.

The 89 full papers presented were carefully reviewed and selected from 195 submissions. The papers included both traditional artificial intelligence methods and soft computing techniques as well as follows:

· Neural Networks and Their Applications

· Fuzzy Systems and Their Applications

· Evolutionary Algorithms and Their Applications

· Artificial Intelligence in Modeling and Simulation

· Computer Vision, Image and Speech Analysis

· Data Mining

· Various Problems of Artificial Intelligence

· Bioinformatics, Biometrics and Medical Applications

Table of Contents

Frontmatter

Neural Networks and Their Applications

Frontmatter
Financial Portfolio Construction for Quantitative Trading Using Deep Learning Technique

Stock portfolio construction is a difficult task which involves the simultaneous consideration of dynamic financial data as well as investment criteria (e.g.: investors required return, risk tolerance, goals, and time frame). The objective of this research is to present a two phase deep learning module to csonstruct a financial stocks portfolio that can be used repeatedly to select the most promising stocks and adjust stocks allocations (namely quantitative trading system). A deep belief network is used to discover the complex regularities among the stocks while a long short-term memory network is used for time series financial data prediction. The proposed deep learning architecture has been tested on the american stock market and has outperformed other known machine learning techniques (support vector machine and random forests) in several prediction accuracy metrices. Furthermore, the results showed that our architecture as a portfolio construction model outperforms three benchmark models with several financial profitability and risk-adjusted metrics.

Rasha Abdel Kawy, Walid M. Abdelmoez, Amin Shoukry
Factor Augmented Artificial Neural Network vs Deep Learning for Forecasting Global Liquidity Dynamics

This paper develops a global liquidity prediction model based on financial and macroeconomic information from different geographical areas. The methodology of the Factor Augmented Artificial Neural Network Model is applied to improve the predictive capacity of liquidity models compared to traditional econometric methodologies. This hybrid methodology based on dynamic factor models and neural networks is compared with Deep Learning methodologies such as Deep Recurrent Convolutional Neural Network and Deep Neural Decision Trees, which has recently shown great results. Our results show the superiority of the precision capacity of Factor Augmented Artificial Neural Network Model over the applied Deep Learning methodology, which demonstrates the importance of data treatment in International Macroeconomics and Finance with techniques from the Vector Autoregressive model. Our conclusions also show the importance of the impact of monetary policy, financial stability, and the real activity of the economy in the behavior of liquidity. This work may be useful for those interest groups in public and macroeconomic policy, showing the potential in the combination of conventional statistical methods with the envelope of Machine Learning techniques.

David Alaminos
Integrate-and-Fire Neurons for Low-Powered Pattern Recognition

Embedded systems acquire information about the real world from sensors and process it to make decisions and/or for transmission. In some situations, the relationship between the data and the decision is complex and/or the amount of data to transmit is large (e.g. in biologgers). Artificial Neural Networks (ANNs) can efficiently detect patterns in the input data which makes them suitable for decision making or compression of information for data transmission. However, ANNs require a substantial amount of energy which reduces the lifetime of battery-powered devices. Therefore, the use of Spiking Neural Networks can improve such systems by providing a way to efficiently process sensory data without being too energy-consuming. In this work, we introduce a low-powered neuron model called Integrate-and-Fire which exploits the charge and discharge properties of the capacitor. Using parallel and series RC circuits, we developed a trainable neuron model that can be expressed in a recurrent form. Finally, we trained its simulation with an artificially generated dataset of dog postures and implemented it as hardware that showed promising energetic properties.

Florian Bacho, Dominique Chu
A New Variant of the GQR Algorithm for Feedforward Neural Networks Training

This paper presents an application of the scaled Givens rotations in the process of feedforward artificial neural networks training. This method bases on the QR decomposition. The paper describes mathematical background that needs to be considered during the application of the scaled Givens rotations in neural networks training. The paper concludes with sample simulation results.

Jarosław Bilski, Bartosz Kowalczyk
Modification of Learning Feedforward Neural Networks with the BP Method

The backpropagation (BP) algorithm is a worldwide used method for learning neural networks. The BP has a low computational load. Unfortunately, this method converges relatively slowly. In this paper a new approach to the backpropagation algorithm is presented. The proposed solution speeds up the BP method by using vector calculations. This modification of the BP algorithm was tested on a few standard examples. The obtained performance of both methods was compared.

Jarosław Bilski, Jacek Smoląg, Patryk Najgebauer
Data-Driven Learning of Feedforward Neural Networks with Different Activation Functions

This work contributes to the development of a new data-driven method (D-DM) of feedforward neural networks (FNNs) learning. This method was proposed recently as a way of improving randomized learning of FNNs by adjusting the network parameters to the target function fluctuations. The method employs logistic sigmoid activation functions for hidden nodes. In this study, we introduce other activation functions, such as bipolar sigmoid, sine function, saturating linear functions, reLU, and softplus. We derive formulas for their parameters, i.e. weights and biases. In the simulation study, we evaluate the performance of FNN data-driven learning with different activation functions. The results indicate that the sigmoid activation functions perform much better than others in the approximation of complex, fluctuated target functions.

Grzegorz Dudek
Time-Domain Signal Synthesis with Style-Based Generative Adversarial Networks Applied to Guided Waves

Data scarcity is a significant problem when it comes to designing machine learning systems for structural health monitoring applications, especially those based around data-hungry algorithms and methods, such as deep learning. Synthetic data generation could potentially alleviate this problem, lowering the number of measurements that need to be acquired in slow and often expensive conventional experiments. Such synthesis can be done by Generative Adversarial Networks, potentially creating unlimited synthetic samples recreating the original data distribution. While most of the research about these networks is centered around using them on image data, they have also been applied to audio waves - going as far as successfully synthesizing human speech. This suggests that these networks should also apply to synthesizing time-domain signals in various fields of structural health monitoring, guided waves in particular, as they are in many ways similar to audio wave signals. This work proposes an adaptation of style-based GAN architecture to time-domain signal generation, and presents its viability for guided waves synthesis, utilizing a database of signals collected in series of pitch-catch experiments on a composite plate.

Mateusz Heesch, Krzysztof Mendrok, Ziemowit Dworakowski
A Comparison of Trend Estimators Under Heteroscedasticity

Trend estimation, i.e. estimating or smoothing a nonlinear function without any independent variables, belongs to important tasks in various applications within signal and image processing, engineering, biomedicine, analysis of economic time series, etc. We are interested in estimating trend under the presence of heteroscedastic errors in the model. So far, there seem no available studies of the performance of robust neural networks or the taut string (stretched string) algorithm under heteroscedasticity. We consider here the Aitken-type model, analogous to known models for linear regression, which take heteroscedasticity into account. Numerical studies with heteroscedastic data possibly contaminated by outliers yield improved results, if the Aitken model is used. The results of robust neural networks turn out to be especially favorable in our examples. On the other hand, the taut string (and especially its robust $$L_1$$ L 1 -version) inclines to overfitting and suffers from heteroscedasticity.

Jan Kalina, Petra Vidnerová, Jan Tichavský
Canine Behavior Interpretation Framework Using Deep Graph Model

Humans have long aspired to understand dog behavior. While research on the Calming signal has achieved substantial progress in understanding dog behavior, it remains an unfamiliar concept to non-expertise. Therefore, in this paper, we introduce a framework for analyzing dog behavior, which defines the interrelationship between dog postures through a graph model without any additional devices but a camera. First of all, our framework classifies the dog posture in frame units, using a machine learning model based on the position information of the dog’s body part in the video captured by the camera. We then analyze dog behavior using graph models that define interrelationships among classified dog postures. We expect that our approach will help non-expertise to understand dog behavior.

Jongmin Lim, Donghee Kim, Kwangsu Kim
Artificial Neural Network Based Empty Container Fleet Forecast

Global trade imbalances and poor, partial and unreliable information about available equipment make the coordination of empty containers a very challenging issue for shipping lines. The cancellation of transport operations once started or the extraordinary repositioning of containers are some of the problems faced by the local shipping agencies. In this paper, we selected the Artificial Neural Networks technique to predict the reception and withdrawal of empty containers in depots to forecast their future stock. To train the predictive models we used the different messages generated along the containers’ shipment journey together with the temporal data related to these events. The evaluation of the models with the test dataset confirmed the possibility of using ANN to predict the number of empty containers in depots.

Joan Meseguer Llopis, Salvador Furio Prunonosa, Miguel Llop Chabrera, Francisco J. Cubas Rodriguez
Efficient Recurrent Neural Network Architecture for Musical Style Transfer

In this paper we present an original method for style transfer between music tracks. We have used a recurrent model consisting of LSTM layers enclosed within an encoder-decoder architecture. In addition, a method for programmatic synthesis of sufficient, paired training datasets using MIDI data was presented. The representation of the data in the form of a real and an imaginary part of short-time Fourier transformation allowed for independent modeling of the music components. The proposed architecture allowed us to improve upon the state of the art solutions in terms of efficiency and range of applications while achieving high precision of the network.

Mateusz Modrzejewski, Konrad Bereda, Przemysław Rokita
Impact of ELM Parameters and Investment Horizon for Currency Exchange Prediction

The foreign exchange market is of the utmost importance for many sectors of the economy, therefore attempts to forecast changes in currency price levels are the research area of many practitioners and theorists. The article aims at examining the impact of settings of various neural network parameters on the results of currency forecasts. The three currency pairs the US dollar, British pound, and Swiss franc to EUR were selected for the analysis. The forecast results for different network settings are examined with three different indicators: forecast error, the ratio of correctly forecasted changes in the course direction and the potential profit generated. The neural network used for the study is Extreme Learning Machine and the forecast horizons taken into account are in the range of one to ten days. The better-quality forecasts based on price levels than on rates of return was shown and good quality forecasts for two out of three currency pairs was obtained in the study. The article also presents the relationship between the results generated by the neural network and the settings of these networks - in particular, the impact of the number of delays on forecast errors and the number of hidden nodes on all three assessment parameters.

Jakub Morkowski
Spectroscopy-Based Prediction of In Vitro Dissolution Profile Using Artificial Neural Networks

In pharmaceutical industry, dissolution testing is part of the target product quality that are essentials in the approval of new products. The prediction of the dissolution profile based on spectroscopic data is an alternative to the current destructive and time-consuming method. Raman and near infrared (NIR) spectroscopies are two complementary methods, that provide information on the physical and chemical properties of the tablets and can help in predicting their dissolution profiles. This work aims to use the information collected by these methods by creating an artificial neural network model that can predict the dissolution profiles of the scanned tablets. The ANN models created used the spectroscopies data along with the measured compression curves as an input to predict the dissolution profiles. It was found that ANN models were able to predict the dissolution profile within the acceptance limit of the f1 and f2 factors.

Mohamed Azouz Mrad, Kristóf Csorba, Dorián László Galata, Zsombor Kristóf Nagy, Brigitta Nagy
Possibilistic Classification Learning Based on Contrastive Loss in Learning Vector Quantizer Networks

Classification in a possibilistic scenario is a kind of multiple class assignments for data. One of the most prominent and interpretable classifier is the learning vector quantization (LVQ) realizing a nearest prototype classifier model. Figuring out the problem of classifying based on possibilistic or probabilistic class labels (assignments) leads to the use of likelihood ratio to organize a sustainable approach. To this end, we start with a special kind of probabilistic LVQ, known as Robust Soft LVQ, and propose a possibilistic extension to pave the way to our new method. Particularly, the proposed possibilistic variant takes positive and negative reasoning known from RSLVQ into account to secure a contrastive learning model in the end. In the paper we will explain the model and give the mathematical justification.

Seyedfakhredin Musavishavazi, Marika Kaden, Thomas Villmann
Convolutional Autoencoder Based Textile Defect Detection Under Unconstrained Setting

Automated visual defect detection on textile products under unconstrained setting is a much sought-after, and at the same time a challenging problem. In general, textile products are structurally complex and highly varied in design, which makes the development of a generalized approach using conventional image processing methods impossible. Deep supervised machine learning models have been very successful on similar problems but cannot be applied in this use-case due to lack of annotated data. This paper demonstrates a novel automated approach which still leverages on the ability of deep learning models to capture complex features on the textured and colored fabric, but in an unsupervised manner. Specifically, deep autoencoders are applied to capture the complex features, which are further processed by image processing techniques like thresholding and blob detection, subsequently leading to detection of defects in the images.

Deepak Nagaraj, Pramod Vadiraja, Oliver Nalbach, Dirk Werth
Clustering-Based Adaptive Self-Organizing Map

We propose an improvement of the Self-Organizing Map (SOM). In our version of SOM, the neighborhood widths of the Best Matching Units (BMUs) are computed on the basis of the data density and scattering in the input data space. The density and scattering are expressed by the values of the inner-cluster variances, which are obtained after the preliminary input data clustering. The experiments conducted on the two real datasets evaluated the proposed approach on the basis of a comparison with the three reference data visualization methods. By reporting the superiority of our technique over the other tested algorithms, we confirmed the effectiveness and accuracy of the introduced solution.

Dominik Olszewski
Applying Machine Learning Techniques to Identify Damaged Potatoes

This paper examines the problem of detecting potatoes with mechanical damage using machine learning techniques.In this article, the authors proposed an algorithm for detecting damaged potato tubers on a conveyor belt that is characterized by speed and accuracy of recognition.The distinctive features of the algorithm are combining the methods of Viola-Jones and the convolutional networks, the application of two complementary classifiers, working in the usual gray color and inverted color. Also, the distinguishing feature is that the identified tubers are processed by the classifiers only once, regardless of the time in front of the video camera.The Viola-Jones method was used to identify individual tubers on the conveyor belt, and the convolutional networks were only used to recognize damaged tubers. Moreover, two complementary networks were used for classification, one of which worked in gray gradation and the other in inverted color.The algorithm was implemented using the OpenCV library in Python. Testing was carried out in conditions close to the conditions of potato storage at vegetable bases.The percentage of properly-recognized damaged tubers was 92,1%.

Aleksey Osipov, Andrey Filimonov, Stanislav Suvorov
Quantifying the Severity of Common Rust in Maize Using Mask R-CNN

The second sustainable development goal defined by the United Nations focuses on achieving food security and supporting sustainable agriculture. This paper focuses on one such initiative contributing to attaining this goal, namely, the identification or prediction of disease in crops. More specifically the paper examines the automated quantification of the severity of common rust in maize. Previous work has focused on using standard image processing algorithms for this problem. This is the first study, to the knowledge of the authors, employing machine learning techniques to determine the severity of common rust disease in maize. Quantifying the severity of common rust is achieved by counting the number of pustules on maize leaves and determining the surface area of the leaf covered by pustules. In this study a Mask R-CNN is used to determine this. Both the standard image processing algorithms and the Mask R-CNN were evaluated on a real-world dataset created from images of maize leaves grown in a greenhouse. The Mask R-CNN was found to outperform the standard image processing algorithms in terms of counting the number of pustules, calculation of the pustule surface area and the average pustule size. These results were found to be statistically significant at a 5% level of significance. One of the challenges with Mask R-CNN is finding suitable parameter values, which is time consuming. Future work will examine automating parameter tuning for the Mask R-CNN.

Nelishia Pillay, Mia Gerber, Katerina Holan, Steven A. Whitham, Dave K. Berger
Federated Learning Model with Augmentation and Samples Exchange Mechanism

The use of intelligent solutions often comes down to the use of already trained classifiers, which is caused by one of their biggest drawbacks. It is the accuracy or effectiveness of artificial intelligence methods, which are algorithms called data-hungry. It means that it depends on the number of samples in the database, and the quality of the classifier could be better if their number is high and the samples are different. In this paper, we propose a solution based on the idea of federated learning in an application for intelligent systems. The proposed solution consists not only in the division of the database among workers but also in the quality of the samples and their possible exchange. Exchanging samples for a particular worker means labeling difficult to classify samples. These samples are used to expand the sets using the generative adversarial network. The mathematical model of a proposal is described, then the experimental results are shown and discussed with the comparison to the classic approach.

Dawid Połap, Gautam Srivastava, Jerry Chun-Wei Lin, Marcin Woźniak
Flexible Data Augmentation in Off-Policy Reinforcement Learning

This paper explores an application of image augmentation in reinforcement learning tasks - a popular regularization technique in the computer vision area. The analysis is based on the model-free off-policy algorithms. As a regularization, we consider the augmentation of the frames that are sampled from the replay buffer of the model. Evaluated augmentation techniques are random changes in image contrast, random shifting, random cutting, and others. Research is done using the environments of the Atari games: Breakout, Space Invaders, Berzerk, Wizard of Wor, Demon Attack. Using augmentations allowed us to obtain results confirming the significant acceleration of the model’s algorithm convergence. We also proposed an adaptive mechanism for selecting the type of augmentation depending on the type of task being performed by the agent.

Alexandra Rak, Alexey Skrynnik, Aleksandr I. Panov
Polynomial Neural Forms Using Feedforward Neural Networks for Solving Differential Equations

Several neural network approaches for solving differential equations employ trial solutions with a feedforward neural network. There are different means to incorporate the trial solution in the construction, for instance one may include them directly in the cost function. Used within the corresponding neural network, the trial solutions define the so-called neural form. Such neural forms represent general, flexible tools by which one may solve various differential equations. In this article we consider time-dependent initial value problems, which requires to set up the trial solution framework adequately.The neural forms presented up to now in the literature for such a setting can be considered as first order polynomials. In this work we propose to extend the polynomial order of the neural forms. The novel construction includes several feedforward neural networks, one for each order. The feedforward neural networks are optimised using a stochastic gradient descent method (ADAM). As a baseline model problem we consider a simple yet stiff ordinary differential equation. In experiments we illuminate some interesting properties of the proposed approach.

Toni Schneidereit, Michael Breuß
Quantum-Hybrid Neural Vector Quantization – A Mathematical Approach

The paper demonstrates how to realize neural vector quantizers by means of quantum computing approaches. Particularly, we consider self-organizing maps and the neural gas vector quantizer for unsupervised learning as well as generalized learning vector quantization for classification learning. We show how quantum computing concepts can be adopted for these algorithms. The respective mathematical framework is explained in detail.

Thomas Villmann, Alexander Engelsberger
A Graphic CNN-LSTM Model for Stock Price Predication

In this paper, we presented a novel model that combines Convolution Neural Network (CNN) and Long Short-term Memory Neural Network (LSTM) for better and accurate stock price prediction. We then developed a model called stock sequence array convolutional LSTM (SACLSTM) that builds both a sequence array of the historical data and leading indicators (i.e., futures and options). This built array is then considered as the input data of the CNN model, thus specific feature vectors via convolutional and pooling layers are then extracted for being the input vector of the LSTM model. Based on this flowchart, the stock price can be better predicted, that can be seen from the conducted experiments in 10 stocks data from USA and Taiwan stock markets. Results also indicated that the designed model is better than the existing models.

Jimmy Ming-Tai Wu, Zhongcui Li, Youcef Djenouri, Dawid Polap, Gautam Srivastava, Jerry Chun-Wei Lin
Applying Convolutional Neural Networks for Stock Market Trends Identification

In this paper we apply a specific type ANNs - convolutional neural networks (CNNs) - to the problem of finding start and endpoints of trends, which are the optimal points for entering and leaving the market. We aim to explore long-term trends, which last several months, not days. The key distinction of our model is that its labels are fully based on expert opinion data. Despite the various models based solely on stock price data, some market experts still argue that traders are able to see hidden opportunities. The labelling was done via the GUI interface, which means that the experts worked directly with images, not numerical data. This fact makes CNN the natural choice of algorithm. The proposed framework requires the sequential interaction of three CNN submodels, which identify the presence of a changepoint in a window, locate it and finally recognize the type of new tendency - upward, downward or flat. These submodels have certain pitfalls, therefore the calibration of their hyperparameters is the main direction of further research. The research addresses such issues as imbalanced datasets and contradicting labels, as well as the need for specific quality metrics to keep up with practical applicability. This is the reduced version of the research, full text will be submitted to arxiv.org.

Ekaterina Zolotareva

Fuzzy Systems and Their Applications

Frontmatter
The Extreme Value Evolving Predictor in Multiple Time Series Learning

This paper extends the evolving fuzzy-rule-based algorithm denoted Extreme Value evolving Predictor (EVeP) to deal with multivariate time series. EVeP offers a statistically well-founded approach to the online definition of the fuzzy granules at the antecedent and consequent parts of evolving fuzzy rules. The interplay established by these granules is used to formulate a regularized multitask learning problem which employs a sparse graph of the structural relationship promoted by the rules. With this multitask strategy, the Takagi-Sugeno consequent terms of the rules are then properly determined. In this extended version, called Extreme Value evolving Predictor in Multiple Time Series Learning (EVeP_MTSL), we propose an approach that resorts to the similarity degree among the time series. The similarity is calculated by the distance correlation statistical measure extracted from a sliding window of data points belonging to the multiple time series. Noticing that each fuzzy rule is part of a specific time series predictor, the new unified model called EVeP_MTSL updates the sparse graph by composing the relationship established by each pair of fuzzy rules (already provided by EVeP) with the similarity degree of their corresponding time series. We are then exploring not only the current interplay of the multiple rules that compose each evolving predictor, but also the current correlation of the multiple time series being simultaneously predicted. Two computational experiments reveal the superior performance of EVeP_MTSL when compared with other contenders devoted to online multivariate time series prediction.

Amanda O. C. Ayres, Fernando J. Von Zuben
Towards Synthetic Multivariate Time Series Generation for Flare Forecasting

One of the limiting factors in training data-driven, rare-event prediction algorithms is the scarcity of the events of interest resulting in an extreme imbalance in the data. There have been many methods introduced in the literature for overcoming this issue; simple data manipulation through undersampling and oversampling, utilizing cost-sensitive learning algorithms, or by generating synthetic data points following the distribution of the existing data. While synthetic data generation has recently received a great deal of attention, there are real challenges involved in doing so for high-dimensional data such as multivariate time series. In this study, we explore the usefulness of the conditional generative adversarial network (CGAN) as a means to perform data-informed oversampling in order to balance a large dataset of multivariate time series. We utilize a flare forecasting benchmark dataset, named SWAN-SF, and design two verification methods to both quantitatively and qualitatively evaluate the similarity between the generated minority and the ground-truth samples. We further assess the quality of the generated samples by training a classical, supervised machine learning algorithm on synthetic data, and testing the trained model on the unseen, real data. The results show that the classifier trained on the data augmented with the synthetic multivariate time series achieves a significant improvement compared with the case where no augmentation is used. The popular flare forecasting evaluation metrics, TSS and HSS, report 20-fold and 5-fold improvements, respectively, indicating the remarkable statistical similarities, and the usefulness of CGAN-based data generation for complicated tasks such as flare forecasting.

Yang Chen, Dustin J. Kempton, Azim Ahmadzadeh, Rafal A. Angryk
The Streaming Approach to Training Restricted Boltzmann Machines

One of the greatest challenges facing researchers of machine learning algorithms nowadays is the desire to minimize the training time of these algorithms. One of the most promising and unexplored structures of the neural network is the Restricted Boltzmann Machine. In this paper, we propose to use the BBTADD algorithm for RBM training. The performance of the algorithm has been illustrated on one of the most popular data sets.

Piotr Duda, Leszek Rutkowski, Piotr Woldan, Patryk Najgebauer
Abrupt Change Detection by the Nonparametric Approach Based on Orthogonal Series Estimates

Many algorithms have been proposed for detection possible deviations and/or narrow changes in the data. A key problem is verification whether the characteristics of information sources have changed. If the change occurred then we would like to know the essence of this change and when or where it happened. Nowadays, well-known methods of mathematical statistics have been successfully applied to address this problem. Recently, a new approach based on nonparametric regression estimation has been proposed. The idea based on Parzen kernel has been studied in depth. This article presents an alternative approach for detecting abrupt change in data based on nonparametric orthogonal series estimation. The proposed method is validated in experiments on noisy data.

Tomasz Gałkowski, Adam Krzyżak
Recommendation System for Signal Processing in SHM

In this article, the recommendation system for processing signals was presented. It contains a database and two fuzzy modules composed within the system. Based on the contextual knowledge provided by the user, collected database, and fuzzy rules, the system suggests processing methods and features. The article presents an evaluation of the proposed system on a two-stage gearbox dataset. The system results are a list of recommended processing methods and extracted features, which allow for a more accurate data classification.

Jakub Gorski, Michal Dziendzikowski, Ziemowit Dworakowski
Monitoring of Changes in Data Stream Distribution Using Convolutional Restricted Boltzmann Machines

In this paper, we propose the Convolutional Restricted Boltzmann Machine (CRBM) as a tool for detecting concept drift in time-varying data streams. Recently, it was demonstrated that the Restricted Boltzmann Machine (RBM) can be successfully applied to this task. A properly learned RBM contains information about the data probability distribution. Trained on one part of the stream it can be used to detect possible changes in the incoming data. In this work we replace the fully-connected layer in the standard RBM with the convolutional layer, composing the CRBM. We show that it is more suitable for the drift detection task regarding the image data. Preliminary experimental results demonstrate the usefulness of the CRBM as a tool for drift detection in data streams with such type of data.

Maciej Jaworski, Leszek Rutkowski, Paweł Staszewski, Patryk Najgebauer
A Proposal for Hybrid Memories Management Exploring Fuzzy-Based Page Migration Policy

This work presents the Intf-HybridMem architecture, a proposal for page migration in hybrid memories using fuzzy systems to support decision making. The fuzzy approach is explored to model the uncertainties of the data access profile and the characteristics of the memory modules. Additionally, the Intf-HybridMem evaluation was carried out, identifying the limit of its accuracy when comparing with the Oracle mechanism.

Lizandro de Souza Oliveira, Rodrigo Costa de Moura, Guilherme Bayer Schneider, Adenauer Correa Yamin, Renata Hax Sander Reiser
A Novel Approach to Determining the Radius of the Neighborhood Required for the DBSCAN Algorithm

Data clustering is one of the most important methods used to discover naturally occurring structures in datasets. One of the most popular clustering algorithms is the Density-Based Spatial Clustering of Applications with Noise (DBSCAN). This algorithm can discover clusters of arbitrary shapes in datasets and thus it has been widely applied in many different applications. However, the DBSCAN requires two input parameters, i.e. the radius of the neighborhood (eps) and the minimum number of points required to form a dense region (MinPts). The right choice of the two parameters is a fundamental issue. In this paper, a new method is proposed to determine the radius parameter. In this approach the distances between each element in the dataset and its k-th nearest neighbor are used, and then in these distances abrupt changes in values are identified. The performance of the new approach has been demonstrated for several different datasets.

Artur Starczewski

Evolutionary Algorithms and Their Applications

Frontmatter
A Learning Automata-Based Approach to Lifetime Optimization in Wireless Sensor Networks

The paper examines the problem of lifetime optimization in Wireless Sensor Networks with an application of a distributed $$\epsilon $$ ϵ -Learning Automaton. The scheme aims to find a global activity schedule maximizing the network’s lifetime while monitoring some target areas with a given measure of requested coverage ratio. The proposed algorithm possesses all the advantages of a localized algorithm, i.e., using only limited knowledge about neighbors, the ability to self-organize in such a way as to prolong the lifetime, and, at the same time, preserving the required coverage ratio of the target field. We present the preliminary results of an experimental study comparing the proposed solution with two centralized algorithms providing an exact (Integer Linear Programming (ILP)) and approximated solution (Genetic Algorithm (GA)) of the studied problem.

Jakub Gąsior, Franciszek Seredyński
Genetic Algorithms in Data Masking: Towards Privacy as a Service

Today’s world is one where the number of publicly stored information and private data is growing exponentially, thus so is the need for more precise and more efficient data protection methods. Data privacy is the field that studies data protection methods as well as privacy models, tools and measures to establish when data is well protected and compliant with privacy requirements. Masking methods are used to perturb a database to permit data analysis while ensuring privacy.This work provides a tool towards privacy as a service. Selecting an appropriate masking method and an appropriate parameterisation is an heuristic process. Our work makes use of genetic algorithms to find a good combination of masking methods and parameters. To do so, a number of solutions (masking methods, parameters) are applied and evaluated, the effectiveness of each solution is measured and well performing solutions are passed on to future generations. Effectiveness of a solution is in terms of information loss and disclosure risk.

Noel Hendrick, Vicenç Torra
Transmission of Genetic Properties in Permutation Problems: Study of Lehmer Code and Inversion Table Encoding

Solution encoding describes the way decision variables are represented. In the case of permutation problems, the classical encoding should ensure that there are no duplicates. During crossover operations, repairs may be carried out to correct or avoid repetitions. The use of indirect encoding aims to define bijections between the classical permutation and a different representation of the decision variables. These encodings are not sensitive to duplicates. However, they lead to a loss of genetic properties during crossbreeding. This paper proposes a study of the impact of this loss both in the space of decision variables and in that of fitness values. We consider two indirect encoding: the Lehmer code and the Inversion table.

Carine Khalil, Wahabou Abdou
Population Management Approaches in the OPn Algorithm

This paper deals with the problem of selecting the population size for the population-based algorithm with dynamic selection of operators (OPn). This research was undertaken to check how population size changes affect the optimization of problems in which both the parameters of the solution and its structure should be selected. Moreover, variants in which the size of the population changes dynamically were considered. The simulations were performed for a small selection/variety of examples of control problems in which the structures and parameters of controllers based on PID systems had to be selected.

Krystian Łapa, Krzysztof Cpałka, Adam Słowik
Harris Hawks Optimisation: Using of an Archive

This paper proposes an enhanced variant of the novel and popular Harris Hawks Optimisation (HHO) method. The original HHO algorithm was studied in many research projects, and a lot of hybrid (cooperative) variants of HHO was proposed. In this research study, an advanced HHO algorithm with an archive of the old solutions is proposed (HHO $$_A$$ A ). The proposed method is experimentally compared with the original HHO algorithm on a set of 22 real-world problems (CEC 2011). The results illustrate the superiority of HHO $$_A$$ A because it outperforms HHO significantly in 20 out of 22 problems, and it is never significantly worse. Four well-known nature-based algorithms were employed to compare the efficiency of the proposed algorithm. HHO $$_A$$ A achieves the best results in overall statistical comparison. A more detailed comparison shows that HHO $$_A$$ A achieves the best results in half real-world problems, and it is never the worst-performing method. A newly employed archive of old solutions significantly increases the performance of the original HHO algorithm.

Petr Bujok
Multiobjective Evolutionary Algorithm for Classifying Cosmic Particles

Classification is the process of predicting the class of objects. It is a type of Supervised Machine Learning, where predefined labels are assigned to objects, based on predetermined criteria. The article presents the idea of the Multiobjective Evolutionary Algorithm (MEA) that supports solving this problem. The proposed MEA uses two optimization criteria: the number of correctly assigned objects and the total distance between objects within the classes. In the process of multiobjective optimization, the algorithm minimizes the number of incorrectly assigned objects and maximizes the consistency of members within classes. The algorithm was tested on a few benchmarks and used to classify cosmic particles, based on their traces detected in Water Cherenkov Detectors (WCD). The results of the experiments suggest that the proposed algorithm takes advantage of the standard single-objective evolutionary algorithm in solving this problem. The algorithm can be also used for solving similar optimization problems.

Krzysztof Pytel
A New Genetic Improvement Operator Based on Frequency Analysis for Genetic Algorithms Applied to Job Shop Scheduling Problem

Many researchers today are using meta-heuristics to treat the class of problems known in the literature as Job Shop Scheduling Problem (JSSP) due to its complexity since it consists of combinatorial problems and it is an NP-Hard computational problem. JSSPs are a resource allocation issue and, to solve its instances, meta-heuristics as Genetic Algorithm (GA) are widely used. Although the GAs present good results in the literature, it is very common for these methods that they are stagnant in solutions that are local optima during their iterations and that have difficulty in adequately exploring the search space. To circumvent these situations, we propose in this work the use of an operator specialized in conducting the GA population to a good exploration: the Genetic Improvement based on Frequency Analysis (GIFA). GIFA makes it possible to manipulate the genetic material of individuals by adding characteristics that are believed to be important, with the proposal of directing some individuals who are lost in the search space to a more favorable subspace without breaking the diversity of the population. The proposed GIFA is evaluated considering two different situations in well-established benchmarks in the specialized JSSP literature and proved to be competitive and robust compared to the methods that represent the state of the art.

Monique Simplicio Viana, Rodrigo Colnago Contreras, Orides Morandin Junior

Artificial Intelligence in Modeling and Simulation

Frontmatter
Decision-Making Problems with Local Extremes: Comparative Study Case

Many MCDA methods have been developed to support the decision-maker in solving complex decision-making problems. Most of them suppose the use of monotonic criteria, such as profit or cost. These methods do not consider the possibility of occurring local extremes in the space of the decision-making problem. Therefore, the question arises about how MCDA methods work when a decision problem consists of non-monotonic criteria.We present a short comparative analysis for four popular MCDA methods, i.e., TOPSIS, VIKOR, PROMETHEE II and COMET. For this purpose, we have used simulations for two different decision-making models. In each case, sets of decision alternatives are generated, then evaluated by the model and selected MCDA methods. The obtained results create rankings from which rank similarity coefficients are calculated. The conducted research shows that the COMET method works better in such conditions than the others, and the VIKOR method does the least well in this task.

Bartłomiej Kizielewicz, Andrii Shekhovtsov, Wojciech Sałabun, Andrzej Piegat
Intelligent Virtual Environments with Assessment of User Experiences

Virtual reality (VR) is a powerful modern medium. The advent of low-cost head-mounted display (HMD) devices made this technology accessible at large and featured VR with possibilities to monitor interactions and user’s motion. However, due to lack of real-time feedback mechanism at present, the level of intelligence for virtual environments is still not sufficient to assist the experience and make group or individual assessments towards VR based applications. In this paper, we present our findings related to the problem of real-time feedback that focus on behavioral data by employing the novel feedback mechanism. Virtual-world coordinates, motions and interactions are tracked and captured in real-time while the user experiences particular application. Captured data is investigated to target the issue of complementing VR applications with features derived from real-time behavioral analysis. In our experiment, we also use collected data and provide a methodology to predict virtual-location by the nonlinear auto-regressive neural network with exogenous inputs (NARX). Results suggest employed neural network model is suitable for performing prediction which can be used to obtain a virtual environment with adaptive intelligence.

Ahmet Köse, Aleksei Tepljakov, Eduard Petlenkov
Intelligent Performance Prediction for Powerlifting

Artificial intelligence methods are successfully applied in many areas where a prediction or classification is needed. An example may also be the forecasting of an athlete’s performance, investigated in this article. Powerlifting is a sport of widespread popularity - more and more training people are serious about the competition and prepare professionally for it. This paper aims to present and analyze the athlete’s deadlift score prediction system based on the previous results and the historical data of other lifters. At first, we use the parametrics to choose an athlete with the results similar to the investigated lifter. We propose the artificial neural network application for smoothing empirical hazard and cumulative distribution functions designated for the failed deadlift attempts. We decided to involve quasi-RBF neural networks – involving the sigmoid function and nonlinear least squares learning algorithm. As a result we get the prediction whether the athlete’s deadlift attempt will be valid or not.

Wojciech Rafajłowicz, Joanna Marszałek
Learning Shape Sensitive Descriptors for Classifying Functional Data

We propose a new method of learning descriptors for constructing classifiers of functional data. These descriptors are moments of a curve derivative, but their learning is based solely on samples of the curve itself. Furthermore, the derivative itself is not directly estimated. This is possible due to the trick of using simultaneously two different bases of a functional space.The advantage of extracting features from the derivative instead of from a curve itself is in raising their sensitivities to a shape of a curve. As expected, this may result in better classification accuracy. The simulation experiments that are based on an augmented real data support this claim, but it is not unconditional. Namely, noticeable improvements can be obtained when an appropriate classifier is selected.

Wojciech Rafajłowicz, Ewaryst Rafajłowicz
An Approach to Contextual Time Series Analysis

This article presents and formally describes an ontology-based approach to domain context formation for time series analysis. Considered the logical representation of the ontology using the descriptive logic ALCHI(D). Also described the experimental results that confirm the correctness and effectiveness of the proposed approach.

Anton Romanov, Aleksey Filippov, Nadezhda Yarushkina
Simulation Analysis of Tunnel Vision Effect in Crowd Evacuation

Excessive cognitive demands, fear, or stress narrow evacuees’ functional fields of view (FFV) in disaster evacuation situations. This tunnel vision hypothesis leads to a new model of evacuee behavior deviating significantly from the previously accepted understanding, and possibly altering conventional evacuation protocol designs. In this study, we analyze the impacts of narrowed vision of evacuees on crowd evacuation efficiency through simulated evacuations. The simulated room to be evacuated included multiple exits, of which only one was correct, as well as a single visual sign designating the correct exit, and an agent found the correct exit via this sign if it was within their FFV. We designed an evacuation decision model for the simulated agents based on herd behavior, including cognitive biases frequently observed during evacuations, to which evacuees were assumed to be subject.

Akira Tsurushima
Backmatter
Metadata
Title
Artificial Intelligence and Soft Computing
Editors
Prof. Leszek Rutkowski
Dr. Rafał Scherer
Marcin Korytkowski
Witold Pedrycz
Prof. Ryszard Tadeusiewicz
Dr. Jacek M. Zurada
Copyright Year
2021
Electronic ISBN
978-3-030-87986-0
Print ISBN
978-3-030-87985-3
DOI
https://doi.org/10.1007/978-3-030-87986-0

Premium Partner