Advances in Computational Intelligence

The 21st century has re-opened the interest of Linguistics on the complexity of natural languages. The equi-complexity dogma –the idea that all languages must be equally complex– has been challenged by a number of researchers that claim that indeed natural languages differ in complexity. In the last fifteen years, challengers of the equi-complexity dogma have proposed many complexity measures that depend on their way of defining complexity. In this paper, we propose a grammatical inference model to measure the relative complexity of languages. The computational tool we introduce is the result of an interdisciplinary study inspired in the process of natural language acquisition.

Grammatical Inference deals with the learning of formal languages from data. Research in this field has mainly reduced the problem of language learning to syntax learning. Taking into account that the theoretical results obtained in Grammatical Inference show that learning formal languages only from syntax is generally hard, in this paper we propose to also take into account contextual information during the language learning process. First, we review works in the area of Artificial Intelligence that use the concept of context, and then, we present the theoretical, algorithmic and practical aspects of our proposal.

Searching for new efficient and exact heuristic optimization algorithms in big search spaces currently remains as an open problem. The search space increases exponentially with the problem size, making impossible to find a solution through a mere blind search. Several heuristic approaches inspired by nature have been adopted as suitable algorithms to solve complex optimization problems in many different areas. Networks of Bio-inspired Processors (NBP) is a formal framework formed of highly parallel and distributed computing models inspired and abstracted by biological evolution. From a theoretical point of view, NBP has been proved broadly to be an efficient solving of NP complete problems. The aim of this paper is to explore the expressive power of NBP to solve hard optimization problems with a big search space, using massively parallel architectures. We use the basic concepts and principles of some metaheuristic approaches to propose an extension of the NBP model, which is able to solve actual problems in the optimization field from a practical point of view.

Networks of Evolutionary Processors (NEP) are a bio-inspired computational model able to solve NP complete problems in an efficient manner. Up to now, the only way to analyze and execute these devices is through hardware and software simulators able to encapsulate the inherent parallelism and the efficiency in their computations. Nowadays, simulators for these models only cover many software applications developed under sequential/parallel architectures over multicore desktop computers or clusters of computers. Most of them, are not able to handle the size of non trivial problems within a massively parallel environment. We consider that cloud computation offers an interesting and promising option to overcome the drawbacks of these solutions. In this paper, we propose a novel parallel distributed architecture to simulate NEPs using on-demand cloud elastic computation. A flexible and extensible simulator is developed in order to demonstrate the suitability and scalability of our architecture with several variants of NEP.

In this paper we describe a possible design to adapt to

Neps of Evolutionary Processors

(NEPs) a general methodogy for simulating natural computers in a distributed way. This methodology was early proposed by other researchers of our group and has proven to be viable and efficient for P-systems (another natural computer with a similar, and even more complex structure than NEPs). We highlight the strcuture, tasks and tools we plan to use in the future implementation of the system. Although several simulators for NEPs are available to the community via internet, almost none of them are designed to be scalable and able to tackle problems of big size.

In this paper we compare the performance of a SSVEP-based BCI spelling application of two different equally sized age groups (five subjects each, ranging from 19 to 27 years and 66 to 70 years). Our results confirm that elderly people may have a slightly deteriorated information transfer rate (ITR). The mean (SD) ITR of the young age group was 27.18 (8.82) bit/min while the elderly people achieved an ITR of 14.42 (6.29) bit/min. The results show that the subject age must be taken into account during the development of a SSVEP-based application.

A brain–computer interface (BCI) is a system that enables people to control an external device by means of their brain activity, without the need of performing muscular activity. BCI systems are normally first tested on a controlled environment before being used in a real, daily scenario. While this is due to security reasons, the conditions that BCI systems users will eventually face in their usual environment may affect their performance in an unforeseen way. In this paper, we try to bridge this gap by presenting a trained BCI user a virtual environment that includes realistic distracting stimuli and testing whether the complexity or the type of such stimuli affects user performance. 11 subjects navigated two virtual environments: a static park and the same one with visual and auditory stimuli simulating typical distractors from a real park. No significant differences were found when using a realistic environment; in other words, the presence of different distracting stimuli did not worsen user performance.

Fatigue at the wheel has been strongly related to car accidents. Traditionally, this phenomenon has been studied in laboratory conditions by means of performance testing. Here, we aimed to improve road safety assessing driver fatigue at the wheel in real scenarios. For this purpose, we have built BioTracker®: a flexible non invasive platform. A smartphone and a microcontroller unit are the core of sysetm. In this paper, we describe BioTracker®, and we present some examples of its implementation.

Brain–Computer Interface (BCI) is particularly relevant as a new way to interact with the outside world for disabled people. Based on P300 event-related potentials (ERPs) BCIs have been frequently used for communication purposes, being the first P300-based BCI paradigm developed by Farwell and Donchin for visual speller. P300-BCI speller studies require a significant attentional demand during sustained long times which could represent fatigue and feeling of increasing workload. The evaluation of workload while using P300-BCI speller requires taking into account the cognitive, emotional and physical state of participant during task. This would help to improve usability of the system. The objective of the study is to evaluate, through objective and subjective measures, three different size of speller in order to analyze effectiveness, cognitive load and user comfort. Three healthy subjects took part in the experiment. The preliminary results suggest that speller size can have different effects on user performance and represent important workload for subjects.

Brain Computer Interface (BCI) technology has achieved limited success outside of laboratory conditions. This technology is hindered by practical considerations of set up, lack of robustness and low Information Transfer Rate (ITR). There are two interfaces in a BCI system: the brain’s interface with the computer and the computer-environment interface, which provides access to applications for the user. Three user services were implemented: control of the smart home, entertainment and communication. These may be accessed through a graphical user interface controlled by a BCI. The paper contrasts the performance of an SSVEP based system with a hybrid BCI comprising eye gaze and muscle response (measured at the scalp). The hybrid developed utilizes the EPOC for recording electrical potential and an EyeTribe gaze tracker; these can be combined to provide more robust interaction with applications. Average ITR for the eye tracker and hybrid approaches (190-200 bpm) are higher than for our SSVEP approach (approx. 15 bpm), for the same applications. The poor performance of our SSVEP system was due to the temporal duration of the stimulation (7s) and partly because not all participants could achieve an accuracy of greater than 50%. The current challenge is the replacement of the scalp recorded muscle component with a reliable user modifiable EEG measure.

Cognitive biometrics aims to user authentication (or identification) by direct measure of electrophysiological signals as response to specific stimuli. In the literature, authentication paradigms for network applications are intended for healthy and independent users with complete control of their muscles. This excludes people with severe motor impairment, such as Brain-computer interface (BCI) users. Conversely, BCIs permit communication with users even in extreme impairment conditions, such as those suffering from locked-in syndrome or in advanced stage of Amyotrophic lateral sclerosis. The downside of BCIs is their very poor performance that, measured in terms of throughput and bit error rate, could lead to impracticable authentication. Specifically, current network applications require users to type long usernames and passwords formed with characters chosen from a large dataset. This forces long BCI sessions that users can not afford due to their heavy cognitive workload. In this paper we present some EEG-based authentication approaches and discuss some relevant aspects that a BCI-based authentication approach should consider for users with severe motor impairment.

This paper proposes and evaluates a filter approach for evolutionary multi-objective feature selection in classification problems with a large number of features. Such classification problems frequently appear in many bioinformatics applications where the number of patterns is smaller than the number of features and thus the curse of dimensionality problem exists. The main contribution of this paper is proposing a set of label-aided utility functions that allows the effective search of the most adequate subset of features through an evolutionary multi-objective optimization scheme. The experimental results have been obtained in a brain-computer interface (BCI) classification task based on LDA classifiers, where the properties of multi-resolution analysis (MRA) for signal analysis in temporal and spectral domains have been used to extract the features from EEG signals. The results from the proposed filter method demonstrate some advantages such as less time consumption and better generalization capabilities with respect to some wrapper-based multi-objective feature selection alternatives.

The task allocation problem is one of the main issues in multi-robot systems. Typical ways to address this problem are based on Swarm Intelligence. One of them is the so-called Response Threshold Method. In the aforementioned method every robot has associated a task response threshold and a task stimuli in such a way that the robot’s probability of executing a certain task depends on both factors. On of the advantage of the aforesaid method is given by the fact that the original problem is treated from a distributed mode which, at the same time, means a very low computational requirements. However, the Response Threshold Method cannot be extended in a natural way to allocate more than two tasks when the theoretical basis is provided by probability theory. Motivated by this fact, this paper leaves the probabilistic approach to the problem and takes a first step towards a possibilistic theoretical approach in order to treat successfully the multi-robot task allocation problem when more than two tasks must be performed. As an example of application, an scenario where each robot task stimuli only depends on the distance between tasks is studied and the convergence of the system to an stable state is shown.

In reactive layers of robotic architectures, behaviors should learn their operation from experience, following the trends of modern intelligence theories. A Case Based Reasoning (CBR) reactive layer could achieve this goal but, as complexity of behaviors increases, the curse of dimensionality arises:too many cases in the behaviors casebases degrade response times so robot’s reactiveness is finally too slow for a good performance. In this work we analyze this problem and propose some improvements in the traditional CBR structure and retrieval phase, at reactive level, to reduce the impact of scalability problems when facing complex behaviors design.

Fitting the skills of the natural vision is an appealing perspective for artificial vision systems, especially in robotics applications where visual perception of the surrounding environment is a key requirement. Focusing on the visual attention dilemma for autonomous visual perception, in this work we propose a model for artificial visual attention combining a statistical foundation of visual saliency and a genetic optimization. The computational issue of our model relies on center-surround statistical features calculations and a nonlinear fusion of different resulting maps. Statistical foundation and bottom-up nature of the proposed model provide as well the advantage to make it usable without needing prior information as a comprehensive solid theoretical basement. The eye-fixation paradigm has been considered as evaluation benchmark providing MIT1003 and Toronto image datasets for experimental validation. The reported experimental results show scores challenging currently best algorithms used in the aforementioned field with faster execution speed of our approach.

This paper presents a novel shared control algorithm for robotized wheelchairs. The proposed algorithm is a new method to extend autonomous navigation techniques into the shared control domain. It reactively combines user’s and robot’s commands into a continuous function that approximates a classic Navigation Function (NF) by weighting input commands with NF constraints. Our approach overcomes the main drawbacks of NFs -calculus complexity and limitations on environment modeling- so it can be used in dynamic unstructured environments. It also benefits from NF properties: convergence to destination, smooth paths and safe navigation. Due to the user’s contribution to control, our function is not strictly a NF, so we call it a pseudo-navigation function (PNF) instead.

The

Autonomous Mini Robot

(AMiRo) is a modular and extensible mini robot platform, designed for scientific research and education. Its decentralized architecture enables to easily add or remove functionalities as required for any application. A well defined physical and electrical interface offers the possibility to design new modules with minimal effort. The open-source software framework for the AMiRo is already growing, since the robot is commonly used for research, education, and competitions. Several demonstrations of the system are given, which present its capabilities. Starting with a fuzzy controller for line following, these demonstrations include remote controlling as well as an implementation of an artificial neural network running on the platform.

Visualization of real world data is a difficult task due to the high-dimensional and the complex structure in real datasets. Scientific data visualization requires a variety of mathematical techniques to transform high-dimensional data sets into simple graphical objects that provide a clearer understanding. In this work a Self-Organizing Spanning Tree is proposed, which is able to learn a tree topology without any prespecified structure. Experimental results are provided to show the good performance with synthetic and real data. Moreover, the proposed self-organizing model is applied to color vector quantization, whose comparative results are provided.

Vertical slot fishways are structures that are placed in rivers to allow fish to avoid obstacles such as dams, hydroelectric plants. Knowing the frequency with which fish go through this type of structures can help to determine their efficiency, as well as know migratory features from species, determine if the fluvial course is healthy or if it is possible to fish with fauna preservation guarantees.

A non-invasive method for fish detection, without the need of direct observation, which uses a laser sensor and computer vision techniques to detect fish, is proposed in this work.

This paper proposes mixture Gaussian neural networks (MGNN) to learn visual features from user specified query image objects or regions for relevance image retrieval. Instead of segmenting query image regions from sample images, relevance feedback feature learning is performed by the proposed MGNN to extract query visual features. After feature learning, the MGNN can be used to measure the appearance difference between the query features and images for image retrieval. The proposed methods were tested on COREL image gallery and the WWW image collections, and testing results were compared with currently leading approaches. From the experimental results, that the extracted and learned query visual features by MGNN can be very close to users’ mind and/or desire, and the closeness is somewhat related to the number of feature leaning iterations. Since any dimensional data can be approximated by mixture Gaussian distributions, thus using MGNN to query and to retrieve similar and/or relevance high dimensional data or images will be a new area of research for future works.

Today, image classification is considered as one of the most important and challenging tasks in computer vision. This paper presents a new method for image classification using Bag Of Visual Words and Local Binary Patterns (LBP). The bag-of-visual-words (BoVW) model has been proven to be very efficient for image classification and image retrieval. However, most proposals directly use local features extracted from an image while ignoring hidden information that could be extracted from an image. To solve this problem, we propose a novel image classification method using information extracted from different channels of the image and the grayscale version of the image. In this way more discriminant information is extracted from the image and as a result the constructed BoVW model gives highly discriminative features that considerably increases the classification performance. In this work we embed features extracted using LBP into BoVW model to construct our proposed scene classification model. The choice of LBP as image feature descriptor is because of the fact that the content of most of the scene images contains textural information so extracting LBP features is a very wise choice compared to other popular image features like Scale Invariant Feature Transform (SIFT) that fails to capture image information in homogeneous areas or textual images. Experiments on Oliva and Torralba (OT) dataset demonstrate the effectiveness of the proposed method.

Weed control in precision agriculture refers to the design of site-specific control treatments according to weed coverage and it is very useful to minimise costs and environmental risks. The crucial component is to provide precise and timely weed maps via weed monitoring. This paper compares different approaches for weed mapping using imagery from Unmanned Aerial Vehicles in sunflower crops. We explore different alternatives, such as object-based analysis, which is a strategy that is spreading rapidly in the field of remote sensing. The usefulness of these approaches is tested by considering support vector machines, one of the most popular machine learning classifiers. The results show that the object-based analysis is more promising than the pixel-based one and demonstrate that both the features related to vegetation indexes and those related to the shape of the objects are meaningful for the problem.

A common symptom of several pathologies is hyperemia, that occurs when a certain tissue has an abnormal hue of red. An increase of blood flow causes the engorgement of blood vessels, which produces the coloration. Hyperemia is an important parameter that specialists take into account when diagnosing diseases such as dry eye syndrome or problems derived from contact lenses wearing. In this work, we propose an automatic methodology to measure the hyperemia level of the bulbar conjunctiva. This methodology emphasizes the transformation from the extracted features to grading scales, using artificial neural networks for the process.

Discrete Wavelet Transform has been widely used in image compression because it is able to compact frequency and spatial localization of image energy into a small fraction of coefficients. In recent years coefficient sign coding has been used to improve image compression. The potential correlation between the sign of a coefficient and the signs of its neighbors opens the possibility to use a sign predictor to improve the image compression process. However, this relationship is not uniform and constant for any image. Typically, image encoders use entropy coding to compact the wavelet coefficients information. This work analyzes the impact of the input parameters over a genetic algorithm that distributes into contexts (sets) the wavelet sign predictors in such a way that the overall aggregate entropy will be reduced as much as possible and a as a consequence higher compression rates can be achieved.

We propose an algorithm for finding a set of texture features characterizing the most homogeneous texture area of an input image. The found set of features is intended for extraction of this segment. The algorithm processes any input images in the absence of any preliminary information about the images and, accordingly, without any learning. The essence of the algorithm is as follows. The image is covered with a number of test windows. In each of them, a degree of texture homogeneity is measured. The test window with maximal degree of homogeneity is determined and a representative patch of pixels is detected. The texture features extracted from the detected representative patch is considered as those that best characterize the most homogeneous texture segment. So, the proposed algorithm facilitates solution of the texture segmentation task by providing a segmentation technique with helpful additional information about the analyzed image. A computer program simulating the algorithm has been created. The program is tested on natural grayscale images.

In this paper a method for improving a tracking algorithm in an augmented reality application is presented. This method addresses several issues to this particular application, like marker-less tracking and color constancy with low quality cameras, or precise tracking with real-time constraints. Due to size restrictions some of the objects are tracked using color information. To improve the quality of the detection, a color selection scheme is proposed to increase color distance between different objects in the scene. Moreover, a new color constancy method based in a diagonal-offset model and k-means is presented. Finally, some real images are used to show the improvement with this new method.

This paper presents a method for image motion estimation for event-based sensors. Accurate and fast image flow estimation still challenges Computer Vision. A new paradigm based on asynchronous event-based data provides an interesting alternative and has shown to provide good estimation at high contrast contours by estimating motion based on very accurate timing. However, these techniques still fail in regions of high-frequency texture. This work presents a simple method for locating those regions, and a novel phase-based method for event sensors that estimates more accurately these regions. Finally, we evaluate and compare our results with other state-of-the-art techniques.

This paper investigates domain generalization: How to use knowledge acquired from related domains and apply it to new domains? Transfer Component Analysis (TCA) learns a shared subspace by minimizing the dissimilarities across domains, while maximally preserving the data variance. We propose Multi-TCA, an extension of TCA to multiple domains as well as Multi-SSTCA, which is an extension of TCA for semi-supervised learning. In addition to the original application of TCA for domain adaptation problems, we show that Multi-TCA can also be applied for domain generalization. Multi-TCA and Multi-SSTCA are evaluated on two publicly available datasets with the tasks of landmine detection and Parkinson telemonitoring. Experimental results demonstrate that Multi-TCA can improve predictive performance on previously unseen domains.

Transfer learning algorithms typically assume that the training data and the test data come from different distribution. It is better at adapting to learn new tasks and concepts more quickly and accurately by exploiting previously gained knowledge. Deep Transfer Learning (DTL) emerged as a new paradigm in transfer learning in which a deep model offer greater flexibility in extracting high-level features. DTL offers selective layer based transference, and it is problem specific. In this paper, we propose the Ensemble of Deep Transfer Learning (EDTL) methodology to reduce the impact of selective layer based transference and provide optimized framework to work for three major transfer learning cases. Empirical results on character, object and biomedical image recognition tasks achieves that the proposed method indicate statistically significant classification accuracy over the other established transfer learning method.

The main objective of the study here presented consists in developing a mathematical forecasting model of the available wind power output for an eight-hour horizon in wind farms that may be affected by inclement meteorological environments where the surface of the wind turbine blades can suffer of ice accumulation. These events may depend on several factors as air temperature, relative humidity, barometric pressure or wind speed, among others. In this way a precise model depending on the referred variables will allow predicting with higher accuracy the available power at the plant when the referred events may occur. A model based in neural networks for the prediction of the available power output of an experimental wind farm has been developed and tested using real data. The proposed model outperforms other professional commercial models.

The interest in dealing with imbalanced datasets has grown in the last years, since they represent many real world scenarios. Different methods that address imbalance problems can be classified into three categories: data sampling, algorithmic modification and cost-sensitive learning. The fundamentals of the last methodology is to assign higher costs to wrong classification classes with the aim of reducing higher cost errors.

In this paper, CO

$$^2$$

RBFN-CS, a cooperative-competitive Radial Basis Function Network algorithm that implements cost-sensitivity is presented. Specifically, a cost parameter is introduced in the training stage of the algorithm. This parameter modifies the learning rate of the weights taking into account the right (or wrong) classification of the instance, the type of class (majority or minority) and the imbalance ratio of the data set.

We present a (TL) framework based on (SDA) for the recognition of immunogold particles. These particles are part of a high-resolution method for the selective localization of biological molecules at the subcellular level only visible through (TEM). Four new datasets were acquired encompassing several thousands of immunogold particles. Due to the particles size (for a particular dataset a particle has a radius of 4 pixels in an image of size 4008

$$\times $$

2670) the annotation of these datasets is extremely time taking. Thereby, we apply a (TL) approach by reusing the learning model that can be used on other datasets containing particles of different (or similar) sizes. In our experimental study we verified that our (TL) framework outperformed the baseline (not involving TL) approach by more than 20% of accuracy on the recognition of immunogold particles.

Feature selection is an important data-preprocessing step that often precedes the classification task. Because of large amount of features in real world applications, feature selection is considered as a hard optimization problem. For such problems, metaheuristics have been shown to be a very promising solving approach. In this work, we propose to use Bee Swarm Optimization (BSO) for feature selection. The proposed algorithm, BSO-FS, is based on the wrapper approach that uses BSO for the generation of feature subsets, and a classifier algorithm to evaluate the solutions. BSO-FS is tested on well-known datasets and its performances are compared with those of recently published methods. Obtained results show that for the majority of datasets, BSO-FS selects efficiently relevant features while improving the classification accuracy.

This paper describes new retrieval algorithms based on heuristic approach in clique-based neural networks introduced by Gripon and Berrou. This associative memory model resembles the well-known Willshaw model with specificity of clustered structure. Several retrieval algorithms exist, for instance, Winners-Take-All and Losers-Kicked-Out. These methods work generally well when the input message suffers reasonable distortions, but the performance drops dramatically in some challenging scenarios because of severe interference. By means of simulations, we show that the proposed heuristic retrieval algorithms are able to significantly mitigate this issue while maintaining biological plausibility to some extent.

This paper starts by revisiting some founding, classical ideas for Neural Networks as Artificial Intelligence devices. The basic functionality of these devices is given by stability related properties such as the gradient-like and other collective qualitative behaviors. These properties can be linked to the structural – connectionist – approach. A version of this approach is offered by the hyperstability theory which is presented in brief (its essentials) in the paper. The hyperstability of an isolated Hopfield neuron and the interconnection of these neurons in hyperstable structures are discussed. It is shown that the so-called “triplet” of neurons has good stability properties with a non-symmetric weight matrix. This suggests new approaches in developing of

Artificial Intelligence

devices based on the triplet interconnection of elementary systems (neurons) in order to obtain new useful emergent collective computational properties.

In this work we will apply some of the Deep Learning models that are currently obtaining state of the art results in several machine learning problems to the prediction of wind energy production. In particular, we will consider both deep, fully connected multilayer perceptrons with appropriate weight initialization, and also convolutional neural networks that can take advantage of the spatial and feature structure of the numerical weather prediction patterns. We will also explore the effects of regularization techniques such as dropout or weight decay and consider how to select the final predictive deep models after analyzing their training evolution.

The paper puts forward an ensemble of state-of-the-art classifiers – support vector machines, neural networks and decision trees – to estimate the length of stay after surgery in patients diagnosed with colorectal cancer. The three paradigms are brought together in order to achieve both a more accurate prediction through a voting scheme and transparency of the discriminative guidelines through visual rules. The results support the theoretical assumptions and are confirmed by the physicians.

Parkinson’s disease (PD) is the second most common neurodegenerative disorder. First appreciable symptoms in PD are those related to an altered movement control. Current PD treatments temporally revert the symptoms, but they do not prevent disease’s progression. At the beginning of the treatment, the antiparkinsonian effect of the medication is very evident and symptoms may completely disappear for hours; however, as disease progresses, motor fluctuations appear. Collecting precise information on the temporal course of fluctuations is essential for tailoring an optimal therapy in PD patients and is one of the main parameters in clinical trials. This paper presents an algorithm for wearable devices to automatically detect patient’s motor fluctuations based on inertial sensors. The algorithm has been evaluated in 7 PD patients at their homes without supervision and performing their usual activities. Results are a mean sensitivity of 99.9% and a mean specificity of 99.9%.

There has been an increasing interest on the analysis of First Person Videos in the last few years due to the spread of low-cost wearable devices. Nevertheless, the understanding of the environment surrounding the wearer is a difficult task with many elements involved. In this work, a method for detecting and mapping the presence of people and crowds around the wearer is presented. Features extracted at the crowd level are used for building a robust representation that can handle the variations and occlusion of people’s visual characteristics inside a crowd. To this aim, convolutional neural networks have been exploited. Results demonstrate that this approach achieves a high accuracy on the recognition of crowds, as well as the possibility of a general interpretation of the context trough the classification of characteristics of the segmented background.

Most people lack awareness and hence understanding about how food-related behavior affects the environment. This commonly results in unsustainable food-related decision making, such as food waste. We propose E-COmate, an augmented bin that measures the weight of food waste through a USB postal scale, a bin and a Raspberry Pi with a Wi-Fi module, and give direct feedback to its users through a tablet by visualizing metaphorical units of the weighted food waste. We intend explore the use of E-COmate in redirecting behavior through transparency, visibility and social influence strategies like social comparison. In this paper, we present our concept, implementation, design rationale and plan of research which we expect to provide insights into the potential of eco-feedback integrated in smart home technology for food sustainability.

Face detection shows a challenging problem in the field of image analysis and computer vision and therefore it has received a great deal of attention over the last few years because of its many applications in various areas. In this paper we propose a new method for face detection using an Extended version of Bag of Visual Words (EBoVW). Two extensions of the original bag of visual words are made in this paper, fist, using Fuzzy C-means instead of K-means clustering and second is, building histogram of words using multiple dictionaries for each image. The performances of the original BoVW model with K-means and the proposed EBoVW are evaluated in terms of Area Under the Curve (AUC) and Equal Error Rate (EER) on MIT CBCL Face dataset which is a very large face dataset. The experimental results show the proposed model achieves very promising results.

Declarative programming has been classically used for solving computational problems regarding AI, knowledge representation and so on. During the last decade, Soft-Computing has emerged as a new application area specially tempting for those new generation declarative languages integrating fuzzy logic into logic programming. In many fuzzy logic programming languages, both program clauses and connective definitions admit a clear declarative, rule-based representation inspired by the well-known logic and functional programming paradigms, respectively. A powerful and promising proposal in this area is represented by the multi-adjoint logic programming approach (for which we have developed the

$$\mathcal F \mathcal L \mathcal O \mathcal P \mathcal E \mathcal R$$

tool), where a set of (logic)

Prolog

-like rules are accompanied with a set of (functional)

Haskell

-like fuzzy connective definitions for manipulating truth degrees beyond the simpler case of

{true,false}

. Since these definitions can be seen as a particular case of equations and/or rewrite rules typically used in functional programming, in this paper we focus on their optimization by reusing some variants of program transformation techniques based on unfolding with a functional taste, which have been largely exploited in this last crisp (not fuzzy) setting. We also show how our method rebounds in the simplification of some computational cost measures we proposed in the past. Our approach is accompanied with some implementation and practical issues in connection with the

$$\mathcal S \mathcal Y \mathcal N \mathcal T \mathcal H$$

and

$$\mathcal F \mathcal L \mathcal O \mathcal P \mathcal E \mathcal R$$

tools and the

fuzzyXPath

application we have developed in the area of the semantic web.

A variety of methods and metrics have been offered so far to measure the extent of similarity among various documents and plagiarism detection systems. However, most of them do not take ambiguity inherent in natural language into account. Therefore, in this paper, a new method taking lexical and semantic features and similarity measures into consideration has been proposed. In the first step, after preprocessing and removing stop word, a text was divided into two parts: general and domain-specific knowledge words. Then, the mixed lexical and semantic fuzzy inference system was designed to assess text similarity. The proposed method was evaluated on Persian paper abstracts of International Conference on e-Learning and e-Teaching (ICELET) Corpus and using IT domain knowledge ontology. The results indicated that the proposed method can achieve a rate of 79% in terms of precision and can detect 83% of the plagiarism cases.

The paper deals with the problem of neural-network based on robust unknown input observer design for the fault diagnosis. Authors review the recent development in the area of robust observers for non-linear discrete-time systems and propose less restrictive procedure for design of the

$${\mathcal {H}_\infty }$$

observer. The approach guaranties simultaneously the unknown input decoupling and the fault estimation. The paper presents an unknown input observer design that reduces to a set of linear matrix inequalities. The final part of the paper presents an illustrative example devoted to fault diagnosis of the wind turbine.

Hierarchical fuzzy systems are proposed to handle the

curse of dimensionality

problem sourced from the use of single fuzzy inference systems with a large number of input parameters. While they are being used in various research problems, each of them is based on a constant hierarchic structure. In this study, this strategy is criticized because it is argued that using a constant hierarchic structure does not guarantee to obtain the most accurate solution for the problem. To observe the effects of structural differences on the prediction performance, experiments are performed on two logical gates by not only utilizing different structures but also different defuzzifiers. In the findings of the experiments, it is proved that the structural variations directly affect the systems’ output, and this differentiation cannot be overcome by changing the defuzzifiers. In addition none of the utilized structures can provide the outputs of equivalent single system. It can be concluded that while applying the hierarchical fuzzy systems on any problem, different structures should be considered to find out the most accurate one that can be constantly utilized for that problem.

The automatic construction of fuzzy system with a large number of input variables involves many difficulties such as large time complexity and getting stuck in a shallow and local minimum. As models to overcome them, the SIRMs (Single Input Rule Modules) and DIRMs (Double Input Rule Modules) models have been proposed. However, they are not always effective in accuracy. In the previous paper, we have proposed the model composed of two phases; the first is a linear transformation of input to intermediate variables and the second is to use SIRMs model. It was shown that the proposed model is superior in accuracy and the number of rules to the conventional models in numerical simulation. In this paper, we will show theoretically that the proposed model is a universal approximator. Further, in order to show the effectiveness of the proposed model, numerical simulation will be performed.

The water influences many areas of society. Energy production, own consumption, and irrigation make use of this resource. Within the electricity production context, the flow forecasting process of the rivers that feed the electricity generation plants is very important for the success of this type of generation. Historically, neural networks have been highlighted in this type of application, in particular, the Multilayer Perceptron. Fuzzy neural networks have also been used for the same purpose. Our goal in this paper is to propose the hybridization of a fuzzy neural network that makes use of Multilayer Perceptron architecture with the Least Squares Method, to the improvement the process of monthly forecast of water flow. The neuro fuzzy network is compared to a Multilayer Perceptron network Classic through experiments and statistical tests. The results showed improvements in predictive process in most cases, suggesting that the new approach has significant potential application.

This article discuss basic concepts of relations between two intuitionistic fuzzy numbers - namely an approach based on deterministic representation, an approach based on intuitionistic fuzzy values, as well as the intuitionistic fuzzy relation. The intuitionistic fuzzy relation definition is based on the extension principle and derived with respect to expected relation properties.

Fuzzy

$$c$$

-means is one of the most well known fuzzy clustering algorithms. It is usually solved using an iterative algorithm. This algorithm does not ensure that the solution is the global optimum. In this paper we study the distribution of values of the objective function of fuzzy

$$c$$

-means.

We also propose a new fuzzy clustering method related to fuzzy

$$c$$

-means. The method presumes that the shape of the membership function is known and can be calculated from the cluster centers, which are the only results of the clustering algorihm.