Search

Purpose This paper aims to describe the use of a meta-learning framework for recommending cost-sensitive classification methods with the aim of answering an important question that arises in machine learning, namely, “Among all the available classification algorithms, and in considering a specific type of data and cost, which is the best algorithm for my problem?” Design/methodology/approach This paper describes the use of a meta-learning framework for recommending cost-sensitive classification methods for the aim of answering an important question that arises in machine learning, namely, “Among all the available classification algorithms, and in considering a specific type of data and cost, which is the best algorithm for my problem?” The framework is based on the idea of applying machine learning techniques to discover knowledge about the performance of different machine learning algorithms. It includes components that repeatedly apply different classification methods on data sets and measures their performance. The characteristics of the data sets, combined with the algorithms and the performance provide the training examples. A decision tree algorithm is applied to the training examples to induce the knowledge, which can then be used to recommend algorithms for new data sets. The paper makes a contribution to both meta-learning and cost-sensitive machine learning approaches. Those both fields are not new, however, building a recommender that recommends the optimal case-sensitive approach for a given data problem is the contribution. The proposed solution is implemented in WEKA and evaluated by applying it on different data sets and comparing the results with existing studies available in the literature. The results show that a developed meta-learning solution produces better results than METAL, a well-known meta-learning system. The developed solution takes the misclassification cost into consideration during the learning process, which is not available in the compared project. Findings The proposed solution is implemented in WEKA and evaluated by applying it to different data sets and comparing the results with existing studies available in the literature. The results show that a developed meta-learning solution produces better results than METAL, a well-known meta-learning system. Originality/value The paper presents a major piece of new information in writing for the first time. Meta-learning work has been done before but this paper presents a new meta-learning framework that is costs sensitive.

Phishing is the most prevalent method of cybercrime that convinces people to provide sensitive information; for instance, account IDs, passwords, and bank details. Emails, instant messages, and phone calls are widely used to launch such cyber-attacks. Despite constant updating of the methods of avoiding such cyber-attacks, the ultimate outcome is currently inadequate. On the other hand, phishing emails have increased exponentially in recent years, which suggests a need for more effective and advanced methods to counter them. Numerous methods have been established to filter phishing emails, but the problem still needs a complete solution. To the best of our knowledge, this is the first survey that focuses on using Natural Language Processing (NLP) and Machine Learning (ML) techniques to detect phishing emails. This study provides an analysis of the numerous state-of-the-art NLP strategies currently in use to identify phishing emails at various stages of the attack, with an emphasis on ML strategies. These approaches are subjected to a comparative assessment and analysis. This gives a sense of the problem, its immediate solution space, and the expected future research directions.

The successful application of deep learning has led to increasing expectations\ud of their use in embedded systems. This in turn, has created the need to find\ud ways of reducing the size of neural networks. Decreasing the size of a neural\ud network requires deciding which weights should be removed without compromising\ud accuracy, which is analogous to the kind of problems addressed by multi-arm\ud bandits. Hence, this paper explores the use of multi-armed bandits for reducing\ud the number of parameters of a neural network. Different multi-armed bandit\ud algorithms, namely e-greedy, win-stay, lose-shift, UCB1, KL-UCB, BayesUCB,\ud UGapEb, Successive Rejects and Thompson sampling are evaluated and their\ud performance compared to existing approaches. The results show that multi-\ud armed bandit pruning methods, especially those based on UCB, outperform other\ud pruning methods.

Phishing is a serious form of online fraud made up of spoofed websites that attempt to gain users’ sensitive information by tricking them into believing that they are visiting a legitimate site. Phishing attacks can be detected many ways, including a user's awareness of fraud protection, blacklisting websites, analyzing the suspected characteristics, or comparing them to recent attempts that followed similar patterns. The purpose of this paper is to create classification models using features extracted from websites to study and classify phishing websites. In order to train the system, we use two datasets consisting of 58,645 and 88,647 URLs labeled as “Phishing” or “Legitimate”. A diverse range of machine learning models such as “XGBOOST, Support Vector Machine (SVM), Random Forest (RF), k-nearest neighbor (KNN), Artificial neural network (ANN), Logistic Regression (LR), Decision tree (DT), and Gaussian naive Bayes (NB)” classifiers are evaluated. ANN provided the best performance with 97.63% accuracy for detecting phishing URLs in experiments. Such a study would be valuable to the scientific community, especially to researchers who work on phishing attack detection and prevention.

The last decade has seen an increase in the attention paid to the development of cost sensitive learning algorithms that\ud aim to minimize misclassification costs while still maintaining accuracy. Most of this attention has been on cost sensitive\ud decision tree learning, while relatively little attention has been paid to assess if it is possible to develop better cost\ud sensitive classifiers based on Bayesian networks. Hence, this paper presents EBNO, an algorithm that utilizes Genetic\ud Algorithms to learn cost sensitive Bayesian networks; where genes are utilized to represent the links between the nodes\ud in Bayesian networks and the expected cost is used as a fitness function. An empirical comparison of the new algorithm\ud has been carried out with respect to: (i) an algorithm that induces cost-insensitive Bayesian networks to provide a base\ud line, (ii) ICET, a well-known algorithm that uses Genetic Algorithms to induce cost-sensitive decision trees, (iii) use of\ud MetaCost to induce cost-sensitive Bayesian networks via bagging (iv) use of AdaBoost to induce cost-sensitive Bayesian\ud networks and (v) use of XGBoost, a gradient boosting algorithm, to induce cost-sensitive decision trees. An empirical\ud evaluation on 28 data sets reveals that EBNO performs well in comparison to the algorithms that produce single\ud interpretable models and performs just as well as algorithms that use bagging and boosting methods.

The advent of price and product comparison sites now makes it even more important to retain customers and identify those that might be at risk of leaving. The use of data mining methods has been widely advocated for predicting customer churn. This paper presents two case studies that utilize decision tree learning methods to develop models for predicting churn for a software company. The first case study aims to predict churn for organizations which currently have an ongoing project, to determine if organizations are likely to continue with other projects. While the second case study presents a more traditional example, where the aim is to predict organizations likely to cease being a subscriber to a service. The case studies include presentation of the accuracy of the models using a standard methodology as well as comparing the results with what happened in practice. Both case studies show the significant savings that can be made, plus potential increase in revenue by using decision tree learning for churn analysis.

This book constitutes the refereed proceedings of the 24th International Conference on Applications of Natural Language to Information Systems, NLDB 2019, held in Salford, UK, in June 2019. The 21 full papers and 16 short papers were carefully reviewed and selected from 75 submissions. The papers are organized in the following topical sections: argumentation mining and applications; deep learning, neural languages and NLP; social media and web analytics; question answering; corpus analysis; semantic web, open linked data, and ontologies; natural language in conceptual modeling; natural language and ubiquitous computing; and big data and business intelligence.

Cost-Sensitive learning has become an increasingly important area that recognizes that real world classification problems need to take the costs of misclassification and accuracy into account. Much work has been done on cost-sensitive decision tree learning, but very little has been done on cost-sensitive Bayesian networks. Although there has been significant research on Bayesian networks there has been relatively little research on learning cost-sensitive Bayesian networks. Hence, this paper explores whether it is possible to develop algorithms that learn cost-sensitive Bayesian networks by taking (i) an indirect approach that changes the data distribution to reflect the costs of misclassification; and (ii) a direct approach that amends an existing accuracy based algorithm for learning Bayesian networks. An empirical comparison of the new approaches is carried out with cost-sensitive decision tree learning algorithms on 33 data sets, and the results show that the new algorithms perform better in terms of misclassification cost and maintaining accuracy.

Sign language is used by approximately 70 million1 people throughout the world, and an automatic tool for interpreting it could make a major impact on communication between those who use it and those who may not understand it.\ud However, computer interpretation of sign language is very difficult given the variability in size, shape and position of the fingers or hands in an image.\ud Hence, this paper explores the applicability of deep learning for interpreting sign language. The paper develops a convolutional neural network aimed at classifying fingerspelling images using both image intensity and depth data.\ud The developed convolutional network is evaluated by applying it to the problem of finger spelling recognition for American Sign Language. The evaluation shows that the developed convolutional network performs better than previous studies and has precision of 82% and recall of 80%. Analysis of the confusion matrix from the evaluation reveals the underlying difficulties of classifying some particular signs which is discussed in the paper.

Decision tree induction is a widely used technique for learning from data which first emerged in the 1980s. In recent years, several authors have noted that in practice, accuracy alone is not adequate, and it has become increasingly important to take into consideration the cost of misclassifying the data. Several authors have developed techniques to induce cost-sensitive decision trees. There are many studies that include pair-wise comparisons of algorithms, but the comparison including many methods has not been conducted in earlier work. This paper aims to remedy this situation by investigating different cost-sensitive decision tree induction algorithms. A survey has identified 30 cost-sensitive decision tree algorithms, which can be organized into ten categories. A representative sample of these algorithms has been implemented and an empirical evaluation has been carried. In addition, an accuracy based look-ahead algorithm has been extended to a new cost-sensitive look-ahead algorithm and also evaluated. The main outcome of the evaluation is that an algorithm based on genetic algorithms, known as ICET, performed better over all the range of experiments thus showing that to make a decision tree cost-sensitive, it is better to include all the different types of costs i.e., cost of obtaining the data and misclassification costs, in the induction of the decision tree.

The last decade has seen a considerable growth in the use of Artificial Intelligence (AI) for operations management with the aim of finding solutions to problems that are increasing in complexity and scale. This paper begins by setting the context for the survey through a historical perspective of OR and AI. An extensive survey of applications of AI techniques for operations management, covering a total of over 1200 papers published from 1995 to 2004 is then presented. The survey utilizes Elsevier's ScienceDirect database as a source. Hence, the survey may not cover all the relevant journals but includes a sufficiently wide range of publications to make it representative of the research in the field. The papers are categorized into four areas of operations management: (a) design, (b) scheduling, (c) process planning and control and (d) quality, maintenance and fault diagnosis. Each of the four areas is categorized in terms of the AI techniques used: genetic algorithms, case-based reasoning, knowledge-based systems, fuzzy logic and hybrid techniques. The trends over the last decade are identified, discussed with respect to expected trends and directions for future work suggested.

Feature selection is a key step in data mining. Unfortunately, there is no single feature selection method that is always the best and the data miner usually has to experiment with different methods using a trial and error approach, which can be time consuming and costly especially with very large datasets. Hence, this research aims to develop a meta learning framework that is able to learn about which feature selection methods work best for a given data set. The framework involves obtaining the characteristics of the data and then running alternative feature selection methods to obtain their performance. The characteristics, methods used and their performance provide the examples which are used by a learner to induce the meta knowledge which can then be applied to predict future performance on unseen data sets.

A significant advance in recent years has been the development of cost-sensitive decision tree learners, recognising that real world classification problems need to take account of costs of misclassification and not just focus on accuracy. The literature contains well over 50 cost-sensitive decision tree induction algorithms, each with varying performance profiles. Obtaining good Bayesian networks can be challenging and hence several algorithms have been proposed for learning their structure and parameters from data. However, most of these algorithms focus on learning Bayesian networks that aim to maximise the accuracy of classifications. Hence an obvious question that arises is whether it is possible to develop cost-sensitive Bayesian networks and whether they would perform better than cost-sensitive decision trees for minimising classification cost? This paper explores this question by developing a new Bayesian network learning algorithm based on changing the data distribution to reflect the costs of misclassification. The proposed method is explored by conducting experiments on over 20 data sets. The results show that this approach produces good results in comparison to more complex cost-sensitive decision tree algorithms.

The mural system was an outcome of a significant effort to develop a support tool for the effective use of a full formal methods development cycle. Experience with it, however, has been limited to a small number of illustrative examples that have been carried out by those closely associated with its development and implementation. This paper aims to remedy this situation by describing the experience of using mural for specifying Dust-Expert, an expert system for the relief venting of dust explosions in chemical processes. The paper begins by summarising the main requirements for Dust-Expert, and then gives a ¯avour of the VDM speci®cation that was formalised using mural.\ud The experience of using mural is described with respect to users' expectations that a formal methods tool should: (i) spot any inconsistencies; (ii) help manage and organise the specifications and allow one to easily add, access, update and delete specifications; (iii) help manage and carry out the refinement process; (iv) help manage and organise theories; (v) help manage and carry out proofs. The paper concludes by highlighting the strengths and weaknesses of mural that could be of interest to those developing the next generation of formal methods development tools.

The validation of data from sensors has become an important issue in the operation and control of modern power plants. One approach is to use knowledge based techniques to detect inconsistencies in measured data. These techniques involve two challenges: real time performance and the use of reasoning methods under uncertainty. This article presents an algorithm for intelligent sensor validation in real time environments. The algorithm utilizes a Bayesian network for the detection of a fault in a set of sensors. This Bayesian network represents the dependencies and independencies among all the sensors. A second Bayesian network isolates the faulty sensor among all the apparent faulty sensors. This isolation is made incrementally, i.e., a probability of failure vector is provided at any time and the quality of the belief measurements increases when more time is spent in the computation. This characteristic makes the algorithm suitable for use in real time environments. An empirical evaluation is presented in the validation of temperature sensors of a gas turbine in a combined cycle plant in Mexico.

One of the main challenges facing systems developers is to design systems with the flexibility and adaptability required to satisfy a professional user's needs. This paper presents a novel architecture and a knowledge representation scheme for a multi-functional adaptive system that support professional engineers that work with established procedures. The system achieves its multi-functionality by using a single knowledge representation scheme based on a set of modular networks. The developed representation is exploited by three operators that enable a user to learn, consult different procedures and solve problems. The representation also facilitates the automatic generation of problems and the identification of user errors. The integration of functionalities through a single representation produces a synergy that results in extra-functionality, flexibility and reduces the amount of development effort required. The system adapts to a user by using simple but efficient user modelling techniques that tailor the tutoring information. A rule-based mechanism is used to propose an agenda and a novel rule simplification algorithm is developed to help an expert to develop the rules for proposing the agenda. A prototype was developed and three civil engineering procedures were implemented to evaluate the architecture through scenarios and an empirical evaluation with real engineers.

Operations management is an area that has recently started to benefit from the use of AI techniques such as expert systems, neural networks and genetic algorithms. These techniques can extend the usefulness of OR modelling and enable new types of decision tasks to be supported by computer-based systems. This paper attempts to review ‘intelligent’ decision support systems and their potential to address some of the problems faced in various areas of operations management. Some useful techniques developed in the field of artificial intelligence are outlined and examples of attempts to use these approaches to support decision making in various areas of operations management are described. Recognising the scale of a complete review of all these areas, emphasis has been given to the most significant and more recent publications.

During the last decade, one important contribution towards requirements engineering has been the advent of formal specification languages. They offer a wellddefined notation that can improve consistency and avoid ambiguity in specifications. However, the process of obtaining formal specifications that are consistent with the requirements is itself a difficult activity. Hence, various researchers are developing systems that aid the transition from informal to formal specifications. The kind of problems tackled and the contributions made by these proposed systems are very diverse. This paper brings these studies together to provide a vision for future architectures that aim to aid the transition from informal to formal specifications. The new architecture, which is based on the strengths of existing studies, tackles a number of key issues in requirements engineering such as identifying ambiguities, incompleteness, and reusability. The paper concludes with a discussion of the research problems that need to be addressed in order to realise the proposed architecture.

The past decade has seen a significant interest on the problem of inducing decision trees that take account of costs of misclassification and costs of acquiring the features used for decision making. This survey identifies over 50 algorithms including approaches that are direct adaptations of accuracy-based methods, use genetic algorithms, use anytime methods and utilize boosting and bagging. The survey brings together these different studies and novel approaches to cost-sensitive decision tree learning, provides a useful taxonomy, a historical timeline of how the field has developed and should provide a useful reference point for future research in this field.

This paper develops a new semantic framework that supports employee performance appraisals, based on inductive logic programming and data mining techniques. The framework is applied to learn a grammar for writing SMART objectives and provide feedback. The paper concludes with an empirical evaluation of the framework which shows promising results.

The validation of data from sensors has become an important issue in the operation and control of modern power plants. One approach is to use knowledge based techniques to detect inconsistencies in measured data. These techniques involve two challenges: real time performance and the use of reasoning methodsunder uncertainty. A promising approach to reasoning about uncertainty is to use Bayesian or causal networks. On their own. “these networks do not address the problem of returning an answer in a limited amount of time. This article therefore develops an extension of Bayesian networks so that inconsistencies can be detected in real-time. The extensions proposed involve the addition of temporal relationships to Bayesian networks and the adoption of a layered architecture to facilitate any time behaviour

In this paper we describe TheMD (The Methods Designer) shell and its use in the development of an expert system, known as Dust-Expert, for the relief of dust explosions. We begin by providing the motivation for the development of TheMD Shell by evaluating a prototype implementation of Dust-Expert using a conventional shell. This evaluation is carried out in terms of the transparency, reliability and extensibility of the system. The lessons learned from this evaluation have led to the development of TheMD Shell. We describe the main features of the shell which include more appropriate schemas, more complete explanation capabilities, a ranking scheme and equation-solving capabilities. We illustrate the use of the schemas and the importance of the equation-solving capabilities by examples from the domain of dust explosions. We also provide a formal specification of the ranking scheme that we adopt. We conclude by highlighting the benefits of our approach in comparison to the usual process of developing knowledge-based systems.

This paper develops a new algorithm for inducing cost-sensitive decision trees that is inspired by the multi-armed bandit problem, in which a player in a casino has to decide which slot machine (bandit) from a selection of slot machines is likely to pay out the most. Game Theory proposes a solution to this multi-armed bandit problem by using a process of exploration and exploitation in which reward is maximized. This paper utilizes these concepts to develop a new algorithm by viewing the rewards as a reduction in costs, and utilizing the exploration and exploitation techniques so that a compromise between decisions based on accuracy and decisions based on costs can be found. The algorithm employs the notion of lever pulls in the multi-armed bandit game to select the attributes during decision tree induction, using a look-ahead methodology to explore potential attributes and exploit the attributes which maximizes the reward. The new algorithm is evaluated on fifteen datasets and compared to six well-known algorithms J48, EG2, MetaCost, AdaCostM1, ICET and ACT. The results obtained show that the new multi-armed based algorithm can produce more cost-effective trees without compromising accuracy. The paper also includes a critical appraisal of the limitations of the new algorithm and proposes avenues for further research.

Much of the early work on data mining concentrated on developing algorithms that focused on classification accuracy. A more challenging and practical problem is to devise algorithms that learn rules or associations that optimize income and take better account of costs. Thus, for example, in marketing, we are interested in identifying investments that produce greatest income given the cost of marketing. In medical diagnosis we need to take account of the costs of tests required in diagnosis, such as blood tests, x-rays and scans. This is an important and growing research area, which has produced interesting work in such areas as associative rule mining, clustering and classification using a variety of different data mining techniques and is used in a wide range of application domains such as business, drug design, cosmetic industry, food industry, pollution detection and financial crisis prediction.

Several authors have studied the problem of inducing decision trees that aim to minimize costs of misclassification and take account of costs of tests. The approaches adopted vary from modifying the information theoretic attribute selection measure used in greedy algorithms such as C4.5 to using methods such as bagging and boosting. This paper presents a new framework, based on game theory, which recognizes that there is a trade-off between the cost of using a test and the misclassification costs. Cost-sensitive learning is viewed as a Multi-Armed Bandit problem, leading to a novel cost-sensitive decision tree algorithm. The new algorithm is evaluated on five data sets and compared to six well known algorithms J48, EG2, MetaCost, AdaCostM1, ICET and ACT. The preliminary results are promising showing that the new multi-armed based algorithm can produce more cost-effective trees without compromising accuracy.

Artificial intelligences techniques such as knowledge based systems, neural networks, fuzzy logic and data mining have been advocated by many researchers and developers as the way to improve many of the software development activities. As with many other disciplines, software development quality improves with the experience, knowledge of the developers, past projects and expertise. Software also evolves as it operates in changing and volatile environments. Hence, there is significant potential for using AI for improving all phases of the software development life cycle. This chapter provides a survey on the use of AI for software engineering that covers the main software development phases and AI methods such as natural language processing techniques, neural networks, genetic algorithms, fuzzy logic, ant colony optimization, and planning methods.

Microflnance provides financial services to low income or poor credit record clients. The 'credit crunch' has led to mainstream lenders tightening their lending policies, resulting in increased financial exclusion. Loan sharks then become an alternative and easy way of borrowing money. However, extremely high interest rates from loan sharks put low income people into worse poverty. Sub-prime lenders play an important role in providing affordable loans to fill the gap between loan sharks and mainstream lenders. All the mainstream lenders have their own loan risk assessment systems, but these systems are either 'in house' or not applicable for giving loans to this marginal group of client. Due to the varying characteristics of this marginal group of clients, sub-prime lenders need to develop their own loan risk assessment system. Although data mining methods have the potential for developing such a risk assessment system, the relative performance of the different data mining methods on such data is not known. Hence, this paper focuses on comparing different data mining methods when applied to loan data for sub-prime lenders.

Employee appraisal systems are widely regarded as fundamental to enhancing the success of organizations. Yet, there is evidence that employees doubt their benefits, organizations find them difficult to implement and their value is questioned. Although computing systems that support appraisals have been developed, their focus remains on recording information, thereby not providing the kind of meaningful and deeper support that could help appraisers and appraises achieve organizational goals.Hence, this paper explores the potential for utilizing sentiment analysis for developing a novel system for supporting appraisals. The paper reviews the challenges facing appraisal processes, summarizes some features of sentiment analysis and then proposes a new semantic framework for appraisal systems. The first steps in developing part of the framework using the GATE system are illustrated with an example showing the potential for using sentiment analysis for assessing whether objectives are SMART.

This volume comprises of selected papers from the 6th IFIP International Conference on Intelligent Information Processing. As the world proceeds quickly into the Information Age, it encounters both successes and challenges, and it is well recognized nowadays that Intelligent Information Processing provides the key to the Information Age and to mastering many of these challenges. Intelligent Information Processing supports the most advanced productive tools that are said to be able to change human life and the world itself. However, the path is never a straight one and every new technology brings with it a spate of new research problems to be tackled by researchers; as a result we are not running out of topics; rather the demand is ever increasing. This conference provides a forum for engineers and scientists in academia, university and industry to present their latest research findings in all aspects of Intelligent Information Processing. The conference received more than 50 papers, of which 35 papers are included in this program as regular papers and 4 as short papers. We are grateful for the dedicated work of both the authors and the referees, and we hope these proceedings will continue to bear fruit over the years to come. All the submitted papers were reviewed by 2 referees. A conference such as this cannot succeed without the help of many individuals who contributed their valuable time and expertise. We want to express our sincere gratitude to the program committee members and referees, who invested many hours for reviews and deliberations. They have provided detailed and constructive review reports that will significantly improve the papers included in the program. We are very grateful to have the sponsorship of the following organizations: IFIP TC12, the University of Salford and the Institute of Computing Technology, Chinese Academy of Sciences .

An exemplar-based model with foundations in Bayesian networks is described. The proposed model utilises two Bayesian networks: one for indexing of categories, and another for identifying exemplars within categories. Learning is incrementally conducted each time a new case is classified. The representation structure dynamically changes each time a new case is classified and a prototypicality function is used as a basis for selecting suitable exemplars. The results of evaluating the model on three datasets are presented.

This paper describes a pragmatic collaborative user modeling approach based on three main components: a multi-functional knowledge base, an open user model, and a set of rules (i.e., the adapting criteria).

An exemplar-based model with foundations in Bayesian networks is described. The proposed model utilises two Bayesian networks: one for indexing of categories, and another for identifying exemplars within categories. Learning is incrementally conducted each time a new case is classified. The representation structure dynamically changes each time a new case is classified and a coverage function is used as a basis for selecting suitable exemplars. Finally, a simple example is given to illustrate the concepts in the model.

A knowledge-based framework called LACEPRO where civil engineers are able to learn, consult and apply established procedures is presented. LACEPRO's knowledge base was designed to be used by three different knowledge operators: the Tutor, the Consultor, and the Expert. Each of these knowledge operators has its own mechanisms to perform their functions over the shared knowledge base. The core of the knowledge base is a set of networks which represent the steps of the engineering procedures. They function as a platform for: teaching procedures, automatic problem generation, diagnosis of user's misunderstandings, problem solving, consultation, and user's navigation through the domain knowledge.

Global competition and rapidly changing customer requirements are demanding increasing changes in manufacturing environments. Enterprises are required to constantly redesign their products and continuously reconfigure their manufacturing systems. Traditional approaches to manufacturing systems do not fully satisfy this new situation. Many authors have proposed that artificial intelligence will bring the flexibility and efficiency needed by manufacturing systems. This paper is a review of artificial intelligence techniques used in manufacturing systems. The paper first defines the components of a simplified intelligent manufacturing systems (IMS), the different Artificial Intelligence (AI) techniques to be considered and then shows how these AI techniques are used for the components of IMS.

Part 7: Computer Vision and Image Understanding; International audience; The author proposed the decoding model of color vision in 1987. International Commission on Illumination (CIE) recommended almost the same symmetric color model for color transform in 2006. For readers to understand the decoding model better, this paper first introduces the decoding model, then uses this model to explain the opponent-process, color evolution, and color blindness pictorially. Recent references on the decoding and reconstruction of colors in the visual cortex induce a new question: what is the decoding algorithm from ganglion cells to the visual cortex? This paper also explains the decoding algorithm. The decoding model is explained as a fuzzy 3–8 decoder. The fuzzy logic used is compatible with Boolean Algebra. The model first obtains the median M of three cones’ outputs B, G, and R, and then obtain three opponent signals by B, G, and R minus M respectively. This model can unify Young and Helmholtz’s tri-pigment theory and Hering’s opponent theory more naturally than the popular zone models. It is symmetrical and compatible with the popular color transform method for computer graphics. The transform from the RGB system to HSV system according to the decoding model is introduced. Several fuzzy 3–8 or 4–16 … fuzzy decoders can be used to construct a Decoding Neural Network (DNN). The decoding algorithm from ganglion cells to the visual cortex can be explained with a two-layer DNN. The reasonability of the decoding model and the potential applications of the DNN are discussed.

Part 1: Machine Learning; International audience; Wikipedia provides rich semantic features e.g., text, link, and category structure. These features can be used to compute semantic similarity (SS) between words or concepts. However, some existing Wikipedia-based SS methods either rely on a single feature or do not incorporate the underlying statistics of different features. We propose novel vector representations of Wikipedia concepts by integrating their multiple semantic features. We utilize the available statistics of these features in Wikipedia to compute their weights. These weights signify the contribution of each feature in similarity evaluation according to its level of importance. The experimental evaluation shows that our new methods obtain better results on SS datasets in comparison with state-of-the-art SS methods.

Part 5: Brain Computer Integration; International audience; How to improve the classification accuracy is a key issue in four-class motor imagery-based brain-computer interface (MI-BCI) systems. In this paper, a method based on phase synchronization analysis and common spatial pattern (CSP) algorithm is proposed. The proposed method embodies the idea of the inverted binary tree, which transforms the multi-class problem into several binary problems. First, the phase locking value (PLV) is calculated as a feature of phase synchronization, then the calculated correlation coefficients of the phase synchronization features are used to construct two pairs of class. Subsequently, we use CSP to extract the features of each class pair and use the linear discriminant analysis (LDA) to classify the test samples and obtain coarse classification results. Finally, the two classes obtained from the coarse classification form a new class pair. We use CSP and LDA to classify the test samples and get the fine classification results. The performance of method is evaluated on BCI Competition IV dataset IIa. The average kappa coefficient of our method is ranked third among the experimental results of the first five contestants. In addition, the classification performance of several subjects is significantly improved. These results show this method is effective for multi-class motor imagery classification.

Part 2: Multi-Agent System; International audience; This work presents AdaptiveSGA, a model for implementing Dynamic Difficulty Scaling through Adaptive Game AI via the Symbiotic Game Agent framework. The use of Dynamic Difficulty Balancing in modern computer games is useful when looking to improve the entertainment value of a game. Moreover, the Symbiotic Game Agent, as a framework, provides flexibility and robustness as a design principle for game agents. The work presented here leverages both the advantages of Adaptive Game AI and Symbiotic Game Agents to implement a robust, efficient and testable model for game difficulty scaling. The model is discussed in detail and is compared to the original Symbiotic Game Agent architecture. Finally, the paper describes how it was applied in simulated soccer. Finally, experimental results, which show that Dynamic Difficulty Balancing was achieved, are briefly analyzed.

Part 1: Machine Learning; International audience; Feature selection is an important part of data mining, image recognition and other fields. The efficiency and accuracy of classification algorithm can be improved by selecting the best feature subset. The classical feature selection technology has some limitations, and heuristic optimization algorithm for feature selection is an alternative method to solve these limitations and find the optimal solution. In this paper, we proposed a Multi Strategy Grey Wolf Optimizer algorithm (MSGWO) based on random guidance, local search and subgroup cooperation strategies for feature selection, which solves the problem that the traditional grey wolf optimizer algorithm (GWO) is easy to fall into local optimization with a single search strategy. Among them, the random guidance strategy can make full use of the random characteristics to enhance the global search ability of the population, and the local search strategy makes grey wolf individuals make full use of the search space around the current best solution, and the subgroup cooperation strategy is very important to balance the global search and local search of the algorithm in the iterative process. MSGWO algorithm cooperates with each other in three strategies to update the location of grey wolf individuals, and enhances the global and local search ability of grey wolf individuals. Experimental results show that MSGWO can quickly find the optimal feature combination and effectively improve the performance of the classification model.

Part 7: Computer Vision and Image Understanding; International audience; Convolutional matrix factorization (ConvMF), which integrates convolutional neural network (CNN) into probabilistic matrix factorization (PMF), has been recently proposed to utilize the contextual information and achieve higher rating prediction accuracy of model-based collaborative filtering (CF) recommender systems. While ConvMF uses max-pooling, which may lose the feature’s location and frequency information. In order to solve this problem, a novel approach with segmented max-pooling (ConvMF-S) has been proposed in this paper. ConvMF-S can extract multiple features and keep their location and frequency information. Experiments show that the rating prediction accuracy has been improved.

Part 5: Brain Computer Integration; International audience; Simultaneous recording of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) have attracted extensive attention and research owing to their high spatial and temporal resolution. However, EEG data are easily influenced by physiological causes, gradient artifact (GA) and ballistocardiogram (BCG) artifact. In this paper, a new blind source separation technique based on dictionary learning is proposed to remove BCG artifact. The dictionary is learned from original data which represents the features of clean EEG signals and BCG artifact. Then, the dictionary atoms are classified according to a list of standards. Finally, clean EEG signals are obtained from the linear combination of the modified dictionary. The proposed method, ICA, AAS, and OBS are tested and compared using simulated data and real simultaneous EEG–fMRI data. The results suggest the efficacy and advantages of the proposed method in the removal of BCG artifacts.

Part 5: Brain Computer Integration; International audience; Recently, deep learning has gained great attention in decoding the neuro-physiological signal. However, which one (classical machine learning or deep learning) has better performance for decoding the functional near-infrared spectroscopy (fNIRS) signal is still lack of full verification. Thus, in this paper, we systematically compared the performance of many classical machine learning methods and deep learning methods in fNIRS data processing for decoding the mental arithmetic task. The classical machine learning methods such as decision tree, linear discriminant analysis (LDA), support vector machine (SVM), K-Nearest Neighbor (KNN) and ensemble methods with strict feature extraction and screening, were used for performance comparison, while the long short-term memory-fully convolutional network (LSTM-FCN) method as a representative of deep leaning methods was applied. Results showed that the classification performance of SVM was the best among the classical machine learning methods, achieving that the average accuracy of the subject-related/unrelated were 91.0% and 83.0%, respectively. Furthermore, the classification accuracy of deep learning was significantly better than that of the involved classical machine learning methods, where the accuracy of deep learning could reach 95.3% with subject-related condition and 97.1% with subject-unrelated condition, respectively. Thus, this paper has totally showed the excellent performance of LSTM-FCN as a representative of deep learning in decoding brain signal from fNIRS dataset, which has outperformed many classical machine learning methods.

Part 7: Computer Vision and Image Understanding; International audience; In today’s world where the number of images is huge and people cannot quickly retrieve the information they need, we urgently need a simpler and more human-friendly way of understanding images, and image captions have emerged. Image caption, as its name suggests, is to analyze and understand image information to generate natural language descriptions of specific images. In recent years, it has been widely used in image-text crossover studies, early infant education, and assisted by disadvantaged groups. And the favor of industry, has produced many excellent research results. At present, the evaluation of image caption is basically based on objective evaluation indicators such as BLUE and CIDEr. It is easy to prevent the generated caption from approaching human language expression. The introduction of GAN idea allows us to use a new method of adversarial training. To evaluate the generated caption, the evaluation module is more natural and comprehensive. Considering the requirements for image fidelity, this topic proposes a GAN-based image description. The Attention mechanism is introduced to improve image fidelity, which makes the generated caption more accurate and more close to human language expression.

Part 6: Pattern Recognition; International audience; With the development of intelligent transportation systems, especially in the context of the comprehensive development and popularization of big data and 5G networks, intelligent transportation signal systems have been experimented and promoted in various countries around the world. As with other big data-based systems, specific attacks pose a threat to the security of big data-based intelligent transportation system systems. Targeting system vulnerabilities, certain simple forms of attack will have a huge impact on signal planning, making Actual traffic is congested. In this article, we first show a specific attack and then add more attack points, analyze the system’s vulnerabilities, and model based on traffic waves and Bayesian predictions, so that the attack points can help the impact is weakened and the traffic can function normally. For experiments, we performed traffic simulation on the VISSIM platform to prove the impact of our attack and further verify the accuracy and effectiveness of the model.

Part 4: Social Computing; International audience; It turns out that the Stochastic Blockmodel (SBM) and its variants can successfully accomplish a variety of tasks, such as discovering community structures. Note that the main limitations are inferencing high time complexity and poor scalability. Our effort is motivated by the goal of harnessing their complementary strengths to develop a scalability SBM for graphs, that also enjoys an efficient inference process and discovery interpretable communities. Unlike traditional SBM that each node is assumed to belong to just one block, we wish to use the node importance to also infer the community membership(s) of each node (as it is one of the goals of SBMs). To this end, we propose a multi-stage maximum likelihood strategy for inferring the latent parameters of adapting the Stochastic Blockmodels to Overlapping Community Detection (OCD-SBM). The intuitive properties to build the model, is more in line with the real-world network to reveal the hidden community structural characteristics. Particularly, this enables inference of not just the node’s membership into communities, but the strength of the membership in each of the communities the node belongs to. Experiments conducted on various datasets verify the effectiveness of our model.

Part 1: Machine Learning; International audience; Salient object detection has recently drawn much attention in computer vision such as image compression and object tracking. Currently, various heuristic computational models have been designed. However, extracting the salient objects with a complex background in the image is still a challenging problem. In this paper, we propose a region merging strategy to extract salient region. Firstly, boundary super-pixels are clustered to generate the initial saliency maps based on the prior knowledge that the image boundaries are mostly background. Next, adjacent regions are merged by sorting the multiple feature values of each region. Finally, we get the final saliency maps by merging adjacent or non-adjacent regions by means of the distance from the region to the image center and the boundary length of overlapping regions. The experiments demonstrate that our method performs favorably on three datasets than state-of-art.

Part 6: Pattern Recognition; International audience; Current voice assistants are offered by large IT companies such as Google, Amazon, Microsoft, Apple or Baidu. The voice assistants include numerous functionalities, which are usually executed centrally in the cloud by the providers. Nevertheless, the providers offer imprecise information on what happens to the input data of the users. Users cannot be sure whether their privacy and data are protected. The central research question is what is currently happening with the voice-based interaction between users and services, and what concepts for configurable data protection by users are conceivable in the future. In this article, we present the survey results obtained by speech assistant users. The results show, in particular, the willingness to pay for individually configurable privacy. The concept for a voice assistant with privacy-awareness is proposed and prototypically implemented.

Part 9: Trust Software; International audience; As underlying infrastructure of cloud computing platform, datacenter is seriously underutilized, however, its operating costs is high. In this paper, we implement virtual machines placement algorithm in CloudSim using constraint programming approach. We first formulate the problem of virtual machines placement in virtualized datacenters as a variant of multi-dimensions bin packing problem, and then exploit constraint solver to solve this problem with the objective of minimizing number of physical machines that host virtual machines. Finally, we compare different virtual placement algorithms for evaluating constraint programming-based virtual machine placement algorithm including the built-in virtual machine placement algorithm in CloudSim and FFD algorithm. The experimental results show that constraint programming-based virtual machines placement algorithm can efficiently reduce the number of physical machines to achieve the goal of reducing datacenter operating costs and improving resource utilization.