Skip to main content

2019 | Buch

Distributed Computing and Artificial Intelligence, 15th International Conference

herausgegeben von: Dr. Fernando De La Prieta, Sigeru Omatu, Antonio Fernández-Caballero

Verlag: Springer International Publishing

Buchreihe : Advances in Intelligent Systems and Computing


Über dieses Buch

The 15th International Symposium on Distributed Computing and Artificial Intelligence 2018 (DCAI 2018) is a forum to present applications of innovative techniques for studying and solving complex problems. The exchange of ideas between scientists and technicians from both the academic and industrial sector is essential to facilitate the development of systems that can meet the ever-increasing demands of today’s society. The present edition brings together past experience, current work and promising future trends associated with distributed computing, artificial intelligence and their application in order to provide efficient solutions to real problems.

This symposium is organized by the University of Castilla-La Mancha, the Osaka Institute of Technology and the University of Salamanca. The present edition was held in Toledo, Spain, from 20th – 22nd June, 2018.


Effects of Switching Costs in Distributed Problem-Solving Systems

In many situations, changing the status quo may induce particular extra costs. Such switching costs are assumed to cause inertia and reduce performance. This paper studies the effects of switching costs in distributed problem-solving systems and, for this, employs an agent-based simulation based on NK fitness landscapes. The results indicate that the complexity of the problem to be solved considerably shapes the effects of switching costs. Depending on the period of time in the search for superior solutions, switching costs may even have beneficial effects in terms of stabilizing the search and increasing the system’s performance.

Friederike Wall
Fire Detection Using DCNN for Assisting Visually Impaired People in IoT Service Environment

In an emergency, such as fire in a building, visually impaired people are prone to danger more than non-impaired people, for they cannot be aware of it quickly. Current fire detection methods such as smoke detector is very slow and unreliable. But by using vision sensor instead, fire can be proven to be detected much faster as shown in our experiments. Previous studies have applied various image processing and machine learning techniques to detect fire, but they usually don’t generalize well because those techniques use hand-crafted features. With the recent advancements in the field of deep learning, this research can be conducted to help solve the problem by using deep learning-based object detector to detect fire. Such approach can learn features automatically, so they can usually generalize well to various scenes. We introduced two object detection models (R1 and R2) with slightly different model’s complexity. R1 can detect fire at 90% average precision and 85% recall at 33 FPS, while R2 has 90% average precision and 61% recall at 50 FPS. The reason why we introduced two models is because we want to have a benchmark comparison as no other research on fire detection with similar techniques exists. We also want to give two model choices when we wish to integrate the model into an IoT platform.

Borasy Kong, Kuoysuong Lim, Jangwoo Kwon
Development of Agent Predicting Werewolf with Deep Learning

In recent years, the development of AI that plays Werewolf attracts attention. This study researches on Werewolf, which is an incomplete information game. In order to create a good game agent, we tried to get unknown information that makes us advantageous in the game. Since Werewolf is communication game, we assumed that there are common strategies or features. For learning such something from enormous game logs, we proposed using LSTM that is a kind of deep learning.

Manami Kondoh, Keinosuke Matsumoto, Naoki Mori
A Comparative Study of Transfer Functions in Binary Evolutionary Algorithms for Single Objective Optimization

Binary versions of evolutionary algorithms have emerged as alternatives to the state of the art methods for optimization in binary search spaces due to their simplicity and inexpensive computational cost. The adaption of such a binary version from an evolutionary algorithm is based on a transfer function that maps a continuous search space to a discrete search space. In an effort to identify the most efficient combination of transfer functions and algorithms, we investigate binary versions of Gravitational Search, Bat Algorithm, and Dragonfly Algorithm along with two families of transfer functions in unimodal and multimodal single objective optimization problems. The results indicate that the incorporation of the v-shaped family of transfer functions in the Binary Bat Algorithm significantly outperforms previous methods in this domain.

Ramit Sawhney, Ravi Shankar, Roopal Jain
Ontology-Based Advertisement Recommendation in Social Networks

With the advent of the Web 2.0 era, a new source of a vast amount of data about users become available. Advertisement recommendation systems are among the applications that can benefit from these data since they can help gain a better understanding of the users’ interests and preferences. However, new challenges emerge from the need to deal with heterogeneous data from disparate sources. Semantic technologies, in general, and ontologies, in particular, have proved effective for knowledge management and data integration. In this work, an ontology-based advertisement recommendation system that leverages the data produced by users in social networking sites is proposed.

Francisco García-Sánchez, José Antonio García-Díaz, Juan Miguel Gómez-Berbís, Rafael Valencia-García
Hand Gesture Detection with Convolutional Neural Networks

In this paper, we present a method for locating and recognizing hand gestures from images, based on Deep Learning. Our goal is to provide an intuitive and accessible way to interact with Computer Vision-based mobile applications aimed to assist visually impaired people (e.g. pointing a finger at an object in a real scene to zoom in for a close-up of the pointed object). Initially, we have defined different hand gestures that can be assigned to different actions. After that, we have created a database containing images corresponding to these gestures. Lastly, this database has been used to train Neural Networks with different topologies (testing different input sizes, weight initialization, and data augmentation process). In our experiments, we have obtained high accuracies both in localization (96%–100%) and in recognition (99.45%) with Networks that are appropriate to be ported to mobile devices.

Samer Alashhab, Antonio-Javier Gallego, Miguel Ángel Lozano
A Genetic Algorithm Model for Slot Allocation Optimization to Brazilian CTOP Approach

The Collaborative Trajectory Options Program (CTOP) makes each airline possible to share its route options to air traffic control center, and so achieve better business goals by reducing strategic operational costs. In Brazil, there are initial efforts to verify the benefits of CTOP implementation to improve the air traffic fluency and financial results. This paper presents a novel approach for Brazilian airspace using Genetic Algorithms to decrease the delay between available slots during CTOP. The slot optimization keeps improving in a safety-separating window of each aircraft en route. The case study presented an reducement about 70% of delay of a certain airline, when used this decision support system by air traffic control authority.

Natan Rodrigues, Leonardo Cruciol, Li Weigang
AllergyLESS. An Intelligent Recommender System to Reduce Exposition Time to Allergens in Smart-Cities

Allergic rhinitis affects between 10% and 30% of the worldwide population. It reduces the quality of life of the individuals and causes losses in the local economy due to absenteeism. AllergyLESS is a recommendation system to solve mobility issues of the citizens by informing them which walking routes minimizes the exposure time to allergens. The system collects air-quality and allergens metrics from wireless pollution stations and open-data sources. In the cases when the pollution stations do not cover the whole area of interest, an ontology for the healthcare domain along with a set of data mining processes are used to forecast the presence of allergens. The system was validated by carrying out a number of controlled simulations of real situations.

José Antonio García-Díaz, José Ángel Noguera-Arnaldos, María Luisa Hernández-Alcaraz, Isabel María Robles-Marín, Francisco García-Sánchez, Rafael Valencia-García
Guided Evolutionary Search for Boolean Networks in the Density Classification Problem

Boolean networks consist of nodes that represent binary variables, which are computed as a function of the values represented by their adjacent nodes. This local processing entails global behaviors, such as the convergence to fixed points, a behavior found in the context of the density classification problem, where the aim is the network’s convergence to a fixed point of the prevailing node value in the initial global configuration of the network; in other words, a global decision is targeted, but according to a constrained, non-global action. Here, we rely on evolutionary searches in order to find rules and network topologies with good performance in the task. All nodes’ neighborhoods are assumed to be defined by non-regular and bidirectional links, and the Boolean function of the network initialized by the local majority rule. Two evolutionary searches are carried out: first, in the space of network topologies, guided by a parameter ($$\omega $$ω) related to the ’small-worldness’ of the networks, and then, in the space of Boolean functions, but constraining the network topologies to the best family identified in the previous experiment. The results clearly make it evident the key and successful role of the $$\omega $$ω parameter in looking for solutions to the task at issue.

Thiago de Mattos, Pedro P. B. de Oliveira
Peculiarity Classification of Flat Finishing Motion Based on Tool Trajectory by Using Self-organizing Maps

The paper proposes an unsupervised classification method for peculiarities of flat finishing motion with an iron file, measured by a 3D stylus. The proposed method extract personal peculiarities based on trajectory of an iron file. The classified peculiarities are used to correct learner’s finishing motions effectively for skill training. In the case of such skill training, the number of classes of peculiarity is unknown. A torus type Self-Organizing Maps is effectively used to classify such unknown number of classes of peculiarity patterns.Experimental results of the classification with measured data of an expert and sixteen learners show effectiveness of the proposed method.

Masaru Teranishi, Shimpei Matsumoto, Hidetoshi Takeno
Classification of Human Body Smell by Learning Vector Quantization

In this paper we consider classification of human body smell using learning vector quantization (LVQ). Smells of human body are classified as sweaty lockerroom smell, middle-aged smell, and age-of-smell. The first one is mainly detected for persons from teenagers to twenties, the second one is for persons from thirties to fifties, and the third one is for persons over fifties. The aim of this paper is to classify smells into three smalles stated above. The sweaty smell is a smell similar to ammonia and isovaleric acid, middle-aged smell is similar to diacetyl, and the age-of-smell is similar to nonenaar. Using a special sampling box, we train the smell sensing data such that each of those smells could be classified into true smell using LVQ. After that, we develop a hardware (Kunkun body) to classify various smell data into each smell.

Sigeru Omatu
A Multi-objective Evolutionary Proposal for Matching Students to Supervisors

In the last few years there has been a growing interest in the use of artificial intelligence to improve different areas of education such as student team formation, learning analytics, intelligent tutoring systems, or the recommendation of learning resources. This paper presents a genetic algorithm that aims to improve the allocation of students to supervisors while taking both the students’ and supervisors’ preferences with regards to research topics, and by providing a balanced allocation for supervisors’ workload. A Pareto optimal genetic algorithm has been designed and tested for the resolution of this problem.

Victor Sanchez-Anguix, Rithin Chalumuri, Vicente Julian
Real-Time Conditional Commitment Logic and Duration Communication Interpreted Systems

We propose a real-time conditional and unconditional commitment logic (RTCTLC) with semantics defined over the duration communication interpreted system – a system with arbitrary integer durations on transitions. The transitions with durations allow us to model different levels of temporal deadlines and to reduce extra verification work resulting from the use of unit measure steps. The whole framework allows us to formally model the behaviour of agents using (conditional, unconditional, and group) commitments and real-time constraints in order to permit reasoning about qualitative and quantitative requirements.

Bożena Woźna-Szcześniak, Ireneusz Szcześniak
Moodsically. Personal Music Management Tool with Automatic Classification of Emotions

It is a fact that music is directly linked to emotions. Various researches study the link between musical characteristics and the feeling produced or even induced. This work shows a web tool that allows the automatically extraction of musical characteristics of songs including the emotional classification and it uses this metadata to manage the user playlist in streaming. The objective of this work has been to contribute to improve a streaming music tool with perceptual characteristics associated with emotions and musical descriptors elements. The tool provides profile management, such as search engine, customizable playlist generation and song’s recommender alignment with emotional elements associated with music characteristics.

Jorge García Vicente, Ana B. Gil, Ana de Luis Reboredo, Diego Sánchez-Moreno, María N. Moreno-García
Arrhythmia Detection Using Convolutional Neural Models

Mostly all works dealing with ECG signal and Convolutional Network approach use 1D CNNs and must train them from scratch, usually applying a signal preprocessing, such as noise reduction, R-peak detection or heartbeat detection. Instead, our approach was focused on demonstrating that effective transfer learning from 2D CNNs can be done using a well-known CNN called AlexNet, that was trained using real images from ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2012. From any temporal signal, it is possible to generate spectral images (spectrograms) than can be analysed by 2D CNN to do the task of extracting automatic features for the classification stage. In this work, the power spectrogram is generated from a randomly ECG segment, so no conditions of signal extraction are applied. After processing the spectrogram with the CNN, its outputs are used as relevant features to be discriminated by a Multi Layer Perceptron (MLP) which classifies them into arrhythmic or normal rhythm segments. The results obtained are in the 90% accuracy range, as good as the state of the art published with 1D CNNs, confirming that transfer learning is a good strategy to develop decision models in signal and image medical tasks.

Jorge Torres Ruiz, Julio David Buldain Pérez, José Ramón Beltrán Blázquez
A Novel Sentence Vector Generation Method Based on Autoencoder and Bi-directional LSTM

Recently, dramatic performance improvement in computing has enabled a breakthrough in machine learning technologies. Against this background, generating distributed representation of discrete symbols such as natural languages and images has attracted considerable interest. In the field of natural language processing, word2vec, a method to generate distributed representations of words is well known and its effectiveness well reported. However, an effective method to generate the distributed representation of sentences and documents has not yet been reported.In this study, we propose a method of generating the distributed representation of sentences by using an autoencoder based on bi-directional long short-term memory (BiLSTM). To obtain the information and findings that necessary to generate effective representations, the computational experiments are carried out.

Kiyohito Fukuda, Naoki Mori, Keinosuke Matsumoto
Recognizing the Order of Four-Scene Comics by Evolutionary Deep Learning

In recent years, comic analysis has become an attractive research topic in the field of artificial intelligence. In this study, we focused on the four-scene comics and applied deep convolutional neural networks (DCNNs) to the data for understanding the order structure. The tuning of the DCNN hyperparameters requires considerable effort. To solve this problem, we propose a novel method called evolutionary deep learning (evoDL) by means of genetic algorithms. The effectiveness of evoDL is confirmed by an experiment conducted to identify structural problems in actual four-scene comics.

Saya Fujino, Naoki Mori, Keinosuke Matsumoto
A Big Data Platform for Industrial Enterprise Asset Value Enablers

The growing ubiquity of IoT, along with bigger steps towards full digitalization in the manufacturing industry, makes it easier to constantly monitor equipment activity and implement predictive maintenance approaches. Big Data solutions are best suited to process the large amounts of data generated through monitorization – additionally, they also allow for processing of unstructured data, such as documents used in not fully-digitalized processes. This paper describes the creation of a small Hadoop cluster, without high-availability, its integration in the InValue architecture and the processes through which it was populated with historical data from a relational warehouse. The degree of parallelization on the data ingestion tasks and its effect on performance were evaluated for the different kinds of datasets that are currently being used for batch data processing.

Alda Canito, Marta Fernandes, Luís Conceição, Isabel Praça, Goreti Marreiros
Estimating the Purpose of Discard in Mahjong to Support Learning for Beginners

It is always difficult for beginners to learn rules of a new game. A game is classified depending on various aspects, for instance, one with perfect or imperfect information. Because of developing the computer program for a game with perfect information, it is significant to focus on a game with imperfect information as the main target for computational research. Especially, Mahjong is one of popular games with imperfect information. From the aspect of the game informatics, we focus on providing a support system for human players. In order to support mahjong beginners, we constructed a system that displays hints and estimates the players’ purpose of discard based on the support vector machine.

Miki Ueno, Daiki Hayakawa, Hitoshi Isahara
A Web-Based Micro-service Architecture for Comparing Parallel Implementations of Dissimilarity Measures

The performance of an application can be significantly improved by using parallelization, as well as by defining micro-services which allow the distribution of the work into several independent tasks. In this paper, we show how a micro-service architecture can be used for developing an efficient and flexible application for the nearest neighbor classification problem. Several dissimilarity measures are compared, in terms of both accuracy and computational time, for sequential as well parallel executions. In addition, a web-based interface was developed in order to facilitate the interaction with the user and easily monitoring the progress of the experiments.

Daniel-Stiven Valencia-Hernández, Ana-Lorena Uribe-Hurtado, Mauricio Orozco-Alzate
A Parallel Application of Matheuristics in Data Envelopment Analysis

Data Envelopment Analysis (DEA) is a non-parametric methodology for estimating technical efficiency and benchmarking. In general, it is desirable that DEA generates the efficient closest targets as benchmarks for each assessed unit. This may be achieved through the application of the Principle of Least Action. However, the mathematical models associated with this principle are based fundamentally on combinatorial NP-hard problems, difficult to be solved. For this reason, this paper uses a parallel matheuristic algorithm, where metaheuristics and exact methods work together to find optimal solutions. Several parallel schemes are used in the algorithm, being possible for them to be configured at different stages of the algorithm. The main intention is to divide the number of problems to be evaluated in equal groups, so that they are resolved in different threads. The DEA problems to be evaluated in this paper are independent of each other, an indispensable requirement for this algorithm. In addition, taking into account that the main algorithm uses exact methods to solve the mathematical problems, different optimization software has been evaluated to compare their performance when executed in parallel. The method is competitive with exact methods, obtaining fitness close to the optimum with low computational time.

Martín González, Jose J. López-Espín, Juan Aparicio, Domingo Giménez
Prediction Market Index by Combining Financial Time-Series Forecasting and Sentiment Analysis Using Soft Computing

In recent years, a lot of research is focusing on predicting real-world outcomes using Social networks data (for example, Twitter Data). Sentiment Analysis of the twitter data thus has become one of the key aspects of making predictions involving human sentiments. Stock market movements are very sensitive and it affects investment of the investors because of this prediction is the main interest of the researchers. Soft computing approaches and nature-inspired computing has a lot of potential in predicting the market movement. In this paper, soft computing techniques are used to predict market trends using sentiments extracted from market data. The results indicate that by selecting suitable neural networks architecture and selecting suitable regression coefficients can improve the overall accuracy and correlation of the predictions. Stock market information people use for investment decisions. Forecasting must be accurate otherwise it will not be effective in the decision. There are techniques like trend based classification, adaptive indicators selection and market trading signals are used in forecasting.

Dinesh Kumar Saini, Kashif Zia, Eimad Abusham
Relaxation Method of Convolutional Neural Networks for Natural Language Processing

Deep learning has developed into one of the most powerful methods in the machine learning field. In particular, convolutional neural networks (CNNs) have been applied not only to image recognition tasks but also to natural language processing (NLP). To reuse older deep learning models, transfer learning techniques have been widely used in the image recognition field. However, there has been little research on transfer learning in NLP. In this paper, we propose a novel transfer learning model based on a relaxation method of CNNs for NLP. The effectiveness of the proposed method is verified using computer simulations, taking a film review score recognition task as an example.

Ryo Iwasaki, Taku Hasegawa, Naoki Mori, Keinosuke Matsumoto
Simple and Linear Bids in Multi-agent Daily Electricity Markets: A Preliminary Report

Variable generation (VG) has several unique characteristics compared to those of traditional thermal and hydro-power plants, notably significant fixed capital costs, but near-zero or zero variable production costs. Increasing the penetration of VG tend to reduce energy prices over time, increase the occurrence of zero or negatively priced periods, and reduce the cleared energy levels of existing plants. This paper presents an overview of an agent-based system, called MATREM, to simulate electricity markets. Special attention is devoted to a case study that aims at analyzing the behavior of a simulated day-ahead market in situations with increasing levels of wind generation, and also comparing market schedules and prices in situations involving either simple and linear bids.

Hugo Algarvio, Fernando Lopes, João Santana
Fault Tolerance in DisCSPs: Several Failures Case

To solve a distributed problem in presence of a failed entity, we have to find a way to accomplish the failed entity tasks. In this paper, we present an approach which guarantees the resolution of DisCSPs in presence of failed agents. This approach is based on local CSPs replication principle: each failed agent local CSP is replicated in another agent which will support it. Obtained results confirm that our approach can solve a DisCSP in presence of failed agents by giving a solution when it exists.

Fadoua Chakchouk, Sylvain Piechowiak, René Mandiau, Julien Vion, Makram Soui, Khaled Ghedira
Computer Vision and the Internet of Things Ecosystem in the Connected Home

An automatic food replenishment system for fridges may help people with cognitive and motor impairments to have a constant food supply at home. More even, sane people may benefit from this system because it is difficult to know accurately and precisely which goods are present in the fridge every day. This system has been a wish and a major challenge for both white good companies and food distributors for decades. It is known that this system requires two things: a sensing module for food stock tracking and another actuating module for food replenishment. The last module can be easily addressed since nowadays there exist many smartphone applications for food delivering, in fact, many food distributors allow their end-users to schedule food replenishment. On the contrary, food stock tracking is not that easy since this requires artificial intelligence to determine not only the different type of goods present in the fridge but also their quantity and quality. In this work, we address the problem of food detection in the fridge by a supervised computer vision algorithm based on Fast Region-based Convolutional Network and an internet of things ecosystem architecture in the connected home for getting high performance on training and deployment of the proposed method. We have tested our method on a data set of images containing sixteen types of goods in the fridge, built with the aid of a fridge-cam. Preliminary results suggest that it is possible to detect different goods in the fridge with good accuracy and that our method may rapidly scale.

Carlos Lopez-Castaño, Carlos Ferrin-Bolaños, Luis Castillo-Ossa
Distributed System Integration Driven by Tests

In complex distributed systems, the integration phase implies a lot of actions due to it is necessary to know how a component interacts with others. Usually, in the system design phase, modules are defined in a hierarchy in order to be easily integrated based on direct dependencies between the modules. That implies a sequential process of integration. In order to accelerate the integration process, agile-inspired integration method has been designed. The method is based in the moment that a unitary test of a component is passed, the dependencies can be started to be tested. The method has been applied in an intelligent system implemented in an indoor drone. First results show that the integration process based on this method is really accelerated, but the coordination between partners and the communication channels have a lot of influence to achieve the process with some minimum quality.

Jose-Luis Poza-Lujan, Juan-Luis Posadas-Yagüe, Stephan Kröner
Coordination Platform for a Swarm of Mobile Robots

In this paper the automatic design of behaviors for a swarm of robots is explored. In order to build behaviors for robots automatically a computational platform is proposed. The proposed platform is composed by three major components. The first component is a description format which allows to specify robot properties, basic behaviors and tasks. The second component is a genetic programming implementation along with a physics-based simulator, this component builds in an automatic way expression trees which represent robot behaviors. The final component is a behaviors allocation module to assign expression trees to real robots. The proposed computational platform is deployed in a experimental manufacturing cell.

John Chavez, Jonatan Gómez, Ernesto Córdoba
Distributed Group Analytical Hierarchical Process by Consensus

The analytical hierarchical process (AHP) is a multi-criteria, decision-making process. This work presents a method to be applied in group decisions (GAHP) using a combination of consensus process and gradient ascent to reach a joint agreement. The GAHP problem is modeled through a multilayer network, where each one of the criteria is negotiated by consensus with the direct neighbors on each layer of the network. Furthermore, each node performs a transversal gradient ascent and corrects the deviations from the personal decision locally. The process locates the optimal global decision, taking into account that this global function is never calculated nor known by any of the agents. If there is not an optimal global decision, but a set of suboptimal choices, agents are automatically divided into different groups that converges into these suboptimal decisions.

M. Rebollo, A. Palomares, C. Carrascosa
Intelligent Flight in Indoor Drones

Currently, drones are one of the most complex control systems. This control covers from the control of the stability of the drone, to the automatic control of the navigation in complex environments. In the case of indoor drones, technological challenges are specific. This paper presents an intelligent control architecture for indoor drones where security is the main axis of the system design. So, a definition of different navigation modes based on security is proposed. The drone must have different navigation modes: manual, reactive, deliberative and intelligent. For indoor navigation it is necessary to know the position of the drone, therefore the system must have a location mode similar to GPS, but that provides better accuracy. For deliberative and intelligent modes, the system must have a map of the environment, as well as a control system that sends the navigation orders to the drone.

Giovanny-Javier Tipantuña-Topanta, Francisco Abad, Ramón Mollá, Jose-Luis Poza-Lujan, Juan-Luis Posadas-Yagüe
Privacy Preserving Expectation Maximization (EM) Clustering Construction

This paper presents a framework for secure Expectation Maximization (EM) clustering construction over partitioned data. It is assumed that data is distributed among several (more than two) parties either horizontally or vertically, such that for mutual benefits all the parties are willing to identify clusters on their data as a whole, but for privacy restrictions, they avoid to share their datasets. To this end, in this study general algorithms based on secure sum is proposed to securely compute the desired criteria in constructing clusters’ scheme.

Mona Hamidi, Mina Sheikhalishahi, Fabio Martinelli
A Framework for Group Decision-Making: Including Cognitive and Affective Aspects in a MCDA Method for Alternatives Rejection

With the evolution of the organizations and technology, Group Decision Support Systems have changed to support decision-makers that cannot be together at the same place and time to make a decision. However, these systems must now be able to support the interaction between decision-makers and provide all the relevant information at the most adequate times. Failing to do so may compromise the success and the acceptance of the system. In this work it is proposed a framework for group decision using a Multiple Criteria Decision Analysis method capable of identify inconsistent assessments done by the decision-maker and identify alternatives that should be rejected by the group of decision-makers. The proposed framework allows to present more relevant information throughout the decision-making process and this way guide decision-makers in the achievement of more consensual and satisfactory decisions.

João Carneiro, Luís Conceição, Diogo Martinho, Goreti Marreiros, Paulo Novais
Domain Identification Through Sentiment Analysis

When dealing with chatbots, domain identification is an important feature to adapt the interactions between user and computer in order to increase the reliability of the communication and, consequently, the audience and decrease its rejection avoiding misunderstandings.In order to adapt to different domains, the writing style will be different for the same author. For example, the same person in the role of a student writes to his professor in a different style than he does for his brother.This article presents a process that uses sentiment analysis to identify the average emotional profile of the communication scenario where the conversation is done. Using Natural Language Processing and Machine Learning techniques, it was possible to obtain an index of 96.21% of correct classifications in the identification of where these communications have occurred only analysing the emotional profile of these texts.

Ricardo Martins, José João Almeida, Pedro Henriques, Paulo Novais
Automatic Music Generation by Deep Learning

This paper presents a model capable of generating and completing musical compositions automatically. The model is based on generative learning paradigms of machine learning and deep learning, such as recurrent neural networks. Related works consider music as a text of a natural language, requiring the network to learn the syntax of the sheet music completely and the dependencies among symbols. This involves a very intense training and may produce overfitting in many cases. This paper contributes with a data preprocessing that eliminates the most complex dependencies allowing the musical content to be abstracted from the syntax. Moreover, a web application based on the trained models is presented. The tool allows inexperienced users to generate automatic music from scratch or from a given fragment of sheet music.

Juan Carlos García, Emilio Serrano
A Novel Hybrid Multi-criteria Decision-Making Model to Solve UA-FLP

The unequal area facility layout problem (UA-FLP) has been addressed by many approaches. Most of them only take quantitative aspects into consideration. In this paper, we will solve UA-FLP using a novel hybrid methodology that joins interactive evolutionary optimization and multi-criteria decision making. I particular, a combination of an interactive genetic algorithm and the analytic hierarchy process (AHP), is proposed. By means of this new approach, it is possible to consider both quantitative and qualitative (using the expert knowledge) criteria in order to reach an acceptable design. Our approach allows the decision maker (DM) to interact with the algorithm, guiding the search process and ranking the criteria that are more relevant in each design solution. In this way, the algorithm is adjusted to the DM’s preferences through his/her subjective evaluations of the representative solutions obtained by a clustering method, and also, to the quantitative criteria. A interesting real-world data set is analysed to empirically probe the robustness of this model. Relevant results are obtained, and interesting conclusions are drawn from the application of this novel intelligent framework.

Laura García-Hernández, L. Salas-Morera, H. Pierreval, Antonio Arauzo-Azofra
Robust Noisy Speech Recognition Using Deep Neural Support Vector Machines

This paper aims to classify noisy sound samples in several daily indoor and outdoor acoustic scenes using an optimized deep neural networks (DNNs). The advantage of a traditional DNNs lies in using at the top layer a softmax activation function which is a logistic regression in order to learn the output label in a multi-class recognition problem. In this paper, we optimize the DNNs by replacing the softmax activation function by a linear support vector machine.In this paper, a novel deep neural networks (DN) using Support Vector Machines (SVM) instead of the multinomial logistic regression is proposed. We have verified the effectiveness of this new method using speech samples from Aurora speech database recorded in noisy conditions. The experimental results obtained with the method DN-SVM demonstrates a significant improvement of the performance with noisy sound samples classification.

Rimah Amami, Dorra Ben Ayed
Preliminary Study of Mobile Device-Based Speech Enhancement System Using Lip-Reading

Inconspicuous speech enhancement system for laryngectomies using lip-reading is proposed to improve the usability and the speech quality. The proposed system uses a tiny camera on mobile phone and recognize the vowel sequences using lip-reading function. Three types of Japanese vowel recognition algorithms using MLP, CNN, and MobileNets, were investigated. 3,000 image datasets for training and testing were prepared from five persons while uttering discrete vowels. Our preliminary experimental result shows that the MobileNets is appropriate for embedding mobile devices in consideration of a performance both recognition accuracy and calculation cost.

Yuta Matsunaga, Kenji Matsui, Yoshihisa Nakatoh, Yumiko O. Kato, Daniel Lopez-Sanchez, Sara Rodriguez, Juan Manuel Corchado
Social Services Diagnosis by Deep Learning

Machine learning and Deep Learning are revolutionizing the field of medicine. These Artificial Intelligence technologies also have the potential to profoundly impact social services. This paper experiments with various architectures of deep artificial neural networks to diagnose cases of chronic social exclusion. The results improve on several metrics previous predictive models based on other machine learning paradigms such as logistic regression or random forests. These models, far from replacing social workers, allow them to be more responsive, efficient, and proactive.

Emilio Serrano, Pedro del Pozo-Jiménez
Combining Image and Non-image Clinical Data: An Infrastructure that Allows Machine Learning Studies in a Hospital Environment

Over the past years Machine Learning and Deep Learning techniques are showing their huge potential in medical research. However, this research is mainly done by using public or private datasets that were created for study purposes. Despite ensuring reproducibility, these datasets need to be constantly updated. In this paper we present an infrastructure that transfers, processes and stores medical image and non-image data in an organized and secure workflow. This infrastructure concept has been tested at a university hospital. XNAT, an extensible open-source imaging informatics software platform was extended to store the non-image data and later feed the Machine Learning models. The resulting infrastructure allowed an easy implementation of a Deep Learning approach for brain tumor segmentation with potential for other medical image research scenarios.

Raphael Espanha, Frank Thiele, Georgy Shakirin, Jens Roggenfelder, Sascha Zeiter, Pantelis Stavrinou, Victor Alves, Michael Perkuhn
GarbMAS: Simulation of the Application of Gamification Techniques to Increase the Amount of Recycled Waste Through a Multi-agent System

The increase in population is increasing the growth of the number of residues. A large amount of this waste can be recycled so that it does not remain in uncontrolled landfills, pollute air, land or water. Although many campaigns and policies of recycling have been developed all-way there is not a total awareness about this problem and a large number of waste does not end up in the right place to be recycled. It is necessary to increase the amount of recycled waste and for this citizen participation is key. It is vital to involve the population in a more active way to recycle through some kind of social benefit. For this reason, GarbMAS proposes a system that generates a motivation for citizen participation by reducing the garbage tax applied by its local government. Thus, the increase in the amount of waste collected to be recycled is expected. GarbMAS employs a multi-agent system that simulates data collection and efficient waste management in cities through gamification techniques to produce the change motivation of citizens and therefore increase participation and recycled quantity. The case study in which the simulation was carried out showed an increase in citizen participation by 34.2% and an increase of 29.4% in the amount of waste collected.

Alfonso González-Briones, Diego Valdeolmillos, Roberto Casado-Vara, Pablo Chamoso, José A. García Coria, Enrique Herrera-Viedma, Juan M. Corchado
A Computational Analysis of Psychopathy Based on a Network-Oriented Modeling Approach

In this paper psychopathy is analysed computationally by creating a temporal-causal network model. The network model was designed using knowledge from Cognitive and Social Neuroscience and simulates the internal neural circuit for moral decision making. Among others, empathy and fear are considered to affect the decision making. This model provides a basis for a virtual agent for simulation-based training or a support application for medical purposes.

Freke W. van Dijk, Jan Treur
An Adaptive Temporal-Causal Network for Representing Changing Opinions on Music Releases

In this paper a temporal-causal network model is introduced representing a shift of opinion about an artist after an album release. Simulation experiments are presented to illustrate the model. Furthermore, mathematical analysis has been done to verify the simulated model and validation by means of an empirical data set and parameter tuning has been addressed as well.

Sarah van Gerwen, Aram van Meurs, Jan Treur
Distributed Computing and Artificial Intelligence, 15th International Conference
herausgegeben von
Dr. Fernando De La Prieta
Sigeru Omatu
Antonio Fernández-Caballero
Electronic ISBN
Print ISBN

Premium Partner