Audio signal classification has been tackled with many machine learning (ML) tools for a diversity of applications. This paper presents a study on Formula One track classification based on the sound of the engine as registered by onboard microphones, including the collection, preprocessing, and curation of the audio data. We carry out a comparative of ML tools applied to this task. Additionally, we evaluate multiple feature extraction methods, including spectrograms, MFCCs, Fourier Transform, Laplace Transform, and Discrete Wavelet Transform assessing their impact on classification performance. The study compares the accuracy of traditional ML models, Logistic Regression, Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), and XGBoost. Experimental results demonstrate the effectiveness of the approach in Formula One track classification classification highlighting the strengths of various models and feature representations.

Retrospective analysis of the data gathered during the COVID-19 pandemic can be used for pandemic preparation in the future. It should be now possible to ascertain if the results of the pandemic policies have been the same across the world. Detecting differences in responses over time can be useful for preparation for future pandemics by posing questions on the causes for these different responses. In this direction, this paper contributes evidence that the pandemic response and the results achieved were not the same everywhere. We consider the multivariate time series of deaths, new people vaccinated, and stringency index for each country. Multivariate Dynamic Time Warping (DTW) allows the elastic matching of this multivariate time series, that results in a similarity measure between countries given by the cost of the elastic matching. Clusters of countries with similar pandemic measures and death time series results can thus be detected by Hierarchical Clustering. We find a cluster composed of some western European countries that is robustly detected under various conditions.

Speech Emotion Recognition (SER) is becoming a key element of speech-based human-computer interfaces, endowing them with some form of empathy towards the emotional status of the human. Transformers have become a central Deep Learning (DL) architecture in natural language processing and signal processing, recently including audio signals for Automatic Speech Recognition (ASR) and SER. A central question addressed in this paper is the achievement of speaker-independent SER systems, i.e. systems that perform independently of a specific training set, enabling their deployment in real-world situations by overcoming the typical limitations of laboratory environments. This paper presents a comprehensive performance evaluation review of transformer architectures that have been proposed to deal with the SER task, carrying out an independent validation at different levels over the most relevant publicly available datasets for validation of SER models. The comprehensive experimental design implemented in this paper provides an accurate picture of the performance achieved by current state-of-the-art transformer models in speaker-independent SER. We have found that most experimental instances reach accuracies below 40% when a model is trained on a dataset and tested on a different one. A speaker-independent evaluation combining up to five datasets and testing on a different one achieves up to 58.85% accuracy. In conclusion, the SER results improved with the aggregation of datasets, indicating that model generalization can be enhanced by extracting data from diverse datasets.

Abstract Artificial Intelligence (AI) has become increasingly integrated into everyday life, with the general population progressively relying on it for even routine tasks. As AI models grow in complexity, precision, and computational power, their energy consumption rises exponentially, raising serious concerns about the sustainability of their widespread adoption. The field of green learning aims to mitigate these concerns by developing energy-efficient AI solutions. In this work, we propose a method to preserve the accuracy of photovoltaic (PV) power generation forecasting while reducing the environmental impact through dataset size reduction. Our approach employs a heuristic strategy that iteratively reduces the training dataset until a cutoff point is reached, balancing predictive accuracy and environmental efficiency. Experimental results using publicly available PV generation datasets demonstrate that the proposed data reduction method decreases training time by up to 17.13%, with only a 1.47% decline in prediction accuracy. These findings highlight the potential of the method to substantially reduce the carbon footprint of AI applications with minimal performance degradation.

The undisputed assumption during the COVID-19 pandemic was that the isolated pathogen SARS-COV-2 was spreading worldwide at lightning speed with the same mortal effects across the world. Under this assumption, synchronized death time series should be expected, however previous results show that COVID-19 mortality time series fall tino very definitive clusters that remain stable under various feature extraction and clustering approaches. Hence, this assumption needs to be reexamined. The research question in this paper is the existence of associations between the intervention measures, namely the social control measures and the massive vaccination programs, and the actual COVID-19 mortality time series. To this end we carry out the spectral clustering of the countries over each of these time series after dimensionality reduction by non-negative matrix factorization (NNMF). Then we look for the maximal intersection between clusters based on each intervention measure and mortality based clusters. The rationale is that clusters of countries based on intervention measures (Vaccination doses, Stringency Index) should greatly overlap with mortality clusters if these measures had any influence on mortality due to COVID-19. Hence, the size of these intersections can be interpreted as a measure of the association between intervention measures and mortality. After exhaustive exploration, the results are disappointing, showing that there is little association between Stringency Index policies and COVID-19 mortality outcomes. Also, we found little association between Vaccination doses and COVID-19 mortality outcomes.

Spatial-temporal land use patterns in urban environments are essential to understanding city dynamics. To uncover these patterns, many researchers have used the digital fingerprints of people’s interaction with urban infrastructure, such as phone calls, facility check-ins, and geolocated social media activity. Despite multiple studies on the detection of land use patterns in urban environments, the need for more consensus on the appropriate metrics to define which set of patterns best describes the dynamics of a city remains a significant limitation. This evaluation is often subjective and depends on the researcher’s in-depth knowledge of the study area, which makes an extensive comparison of multiple cities difficult. This paper introduces a novel set of metrics to determine the patterns that best represent urban activity, diminishing subjective interpretations. Our methodology, which tests our metrics on land use patterns obtained from a dynamic topic model, is a fresh approach to the field. To apply our methodology, we use a dataset of human urban activities collected over 17 years in cities with more than 1 million inhabitants or country capitals. Our results demonstrate that these metrics are a starting point for understanding, analyzing, and choosing the land use patterns that best describe the dynamics and use given to urban space. The practical implications of our research are significant, as it can guide decision-making processes and contribute to the sustainable development of urban areas. However, it is important to highlight that there is still work to be done to reach a consensus on the optimal metrics to evaluate these patterns.

Simion Zoo is a Reinforcement Learning (RL) workbench developed for training of novel users that can be deployed over computer farms allowing extensive experimentation over distributed resources. In this paper, we present this software platform and share some of the insights gained during the development. This workbench provides a complete set of tools to design, run, carry out statistical analysis of the results, report preparation, and have qualitative visual assessment of the simulation evolution of continuous RL control experiments. The main features that set apart Simion Zoo from other software packages for introduction to RL experimentation are its easy-to-use GUI, its support for distributed execution including deployment over graphics processing units (GPUs), and the possibility to explore concurrently the RL hyper-parameter space, which is key to successful RL experimentation.

Electroencephalography (EEG) sensors are flexible and non-invasive sensoring devices for the measurement of electrical brain activity which is extensively used in some areas of clinical practice and psychological/psychiatric research, such as epilepsy, sleep, emotion, and brain computer interfaces. Although EEG sensor do not provide actual brain localizations of the activity sources, they allow to study brain functional connectivity. In this paper we review current application of a specific family of computational methods, the Graph Neural Networks (GNN) to the analysis of EEG data. GNNs appear to be well suited to EEG data modeling as they deal with signals whose domain is defined by a graph instead of a regular lattice in Euclidean space. Readings of EEG electrodes fall in this category, hence the increasing research activity on the application of GNNs to EEG data.

User emotional status recognition is becoming a key feature in advanced Human Computer Interfaces (HCI). A key source of emotional information is the spoken expression, which may be part of the interaction between the human and the machine. Speech emotion recognition (SER) is a very active area of research that involves the application of current machine learning and neural networks tools. This ongoing review covers recent and classical approaches to SER reported in the literature.

Abstract Aging population in many developed countries, moves the issue of healthy aging at the forefront of the political, scientific and technological concerns. Delirium is a multifactorial disorder that is highly prevalent in hospitalized elderly people that causes complications in the patient care and increases mortality at the hospital and soon after discharge. Early diagnostics would allow improved treatment and prevention for a syndrome that requires very personalized treatment. This paper deals with machine learning based prediction of delirium at hospital admittance as a computer aided diagnostic tool, as well as with the identification of risk factors by means of the variable importance computed by the classifier model building approaches. We achieve almost 0.80 classification accuracy, which is encourages further exploration of improved classifier models. Exploration of variable importance shows that frailty, dementia and some pharmacological factors are relevant risk factors for delirium at hospital admittance.

In recent years, speech emotion recognition (SER) has emerged as one of the most active human-machine interaction research areas. Innovative electronic devices, services and applications are increasingly aiming to check the user emotional state either to issue alerts under some predefined conditions or to adapt the system responses to the user emotions. Voice expression is a very rich and noninvasive source of information for emotion assessment. This paper presents a novel SER approach based on that is a hybrid of a time-distributed convolutional neural network (TD-CNN) and a long short-term memory (LSTM) network. Mel-frequency log-power spectrograms (MFLPSs) extracted from audio recordings are parsed by a sliding window that selects the input for the TD-CNN. The TD-CNN transforms the input image data into a sequence of high-level features that are feed to the LSTM, which carries out the overall signal interpretation. In order to reduce overfitting, the MFLPS representation allows innovative image data augmentation techniques that have no immediate equivalent on the original audio signal. Validation of the proposed hybrid architecture achieves an average recognition accuracy of 73.98% on the most widely and hardest publicly distributed database for SER benchmarking. A permutation test confirms that this result is significantly different from random classification ([Formula: see text]). The proposed architecture outperforms state-of-the-art deep learning models as well as conventional machine learning techniques evaluated on the same database trying to identify the same number of emotions.

Computational Ethology studies focused on human beings is usually referred as Human Activity Recognition (HAR). Specifically, this paper belongs to a line of work on the identification of broad cognitive activities that users carry out with computers. The keystone of this kind of systems is the noninvasive detection of the subject’s gaze fixations in selected display areas. Noninvasiveness is ensured by using the conventional laptop cameras without additional illumination or tracking devices. The gaze ethograms, composed as sequences of gaze fixations, are the basis to identify the user activities. To determine the gaze fixation display areas with the highest accuracy, this paper explores the use of a transfer learning approach applied to several well-known deep learning network (DLN) architectures whose input is the eye area extracted from the face image,and output is the identification of the gaze fixation area in the computer screen. Two different datasets are created and used in the validation experiments. We report encouraging results that may allow the general use of the system.

The control of Variable-Speed Wind-Turbines (VSWT) extracting electrical power from the wind kinetic energy are composed of subsystems that need to be controlled jointly, namely the blade pitch and the generator torque controllers. Previous state of the art approaches decompose the joint control problem into independent control subproblems, each with its own control subgoal, carrying out separately the design and tuning of a parameterized controller for each subproblem. Such approaches neglect interactions among subsystems which can introduce significant effects. This paper applies Actor-Critic Reinforcement Learning (ACRL) for the joint control problem as a whole, carrying out the simultaneous control parameter optimization of both subsystems without neglecting their interactions, aiming for a globally optimal control of the whole system. The innovative control architecture uses an augmented input space so that the parameters can be fine-tuned for each working condition. Validation results conducted on simulation experiments using the state-of-the-art OpenFAST simulator show a significant efficiency improvement relative to the best state of the art controllers used as benchmarks, up to a 22% improvement in the average power error performance after ACRL training.

Abstract Understanding at microscopic level the generation of contents in an online social network (OSN) is highly desirable for an improved management of the OSN and the prevention of undesirable phenomena, such as online harassment. Content generation, i.e., the decision to post a contributed content in the OSN, can be modeled by neurophysiological approaches on the basis of unbiased semantic analysis of the contents already published in the OSN. This paper proposes a neuro-semantic model composed of (1) an extended leaky competing accumulator (ELCA) as the neural architecture implementing the user concurrent decision process to generate content in a conversation thread of a virtual community of practice, and (2) a semantic modeling based on the topic analysis carried out by a latent Dirichlet allocation (LDA) of both users and conversation threads. We use the similarity between the user and thread semantic representations to built up the model of the interest of the user in the thread contents as the stimulus to contribute content in the thread. The semantic interest of users in discussion threads are the external inputs for the ELCA, i.e., the external value assigned to each choice.. We demonstrate the approach on a dataset extracted from a real life web forum devoted to fans of tinkering with musical instruments and related devices. The neuro-semantic model achieves high performance predicting the content posting decisions (average F score 0.61) improving greatly over well known machine learning approaches, namely random forest and support vector machines (average F scores 0.19 and 0.21).

Lane markings are mymargin a key element for Autonomous Driving. The generation of high definition maps and ground-truth data require extensive manual labor. In this paper, we present an efficient and robust method for the offline annotation of lane markings, using low-density LIDAR point clouds and odometry information. The odometry is used to accumulate the scans and to process them using blocks following the trajectory of the vehicle. At each block, candidate lane marking points are detected by generating virtual scan-lines and applying a dynamically optimized filter function to the LIDAR intensity values. The lane markings are tracked block wise, and their width is estimated and classified as either solid or dashed. The results are lists of connected 3D points that represent the different lane markings. The accuracy of the proposed method was tested against manually labeled recordings. A novel evaluation methodology focused on the lateral precision of detections is presented. Moreover, a web user interface was used to load the produced annotations, achieving a reduction of 60% in the annotation time, as compared to a fully manual baseline.

Abstract This study argues the difference between security and privacy and outlines the concept of Privacy Debt as a new Technical Debt. Privacy is gaining momentum in any software system due to mandatory compliance with respect to laws and regulations. There are several types of technical debts within the umbrella of software engineering, and most of them arise during different phases of software development. Several research studies have been focussed on highlighting different types of technical debts. However, authors introduce Privacy Debt as a particular technical debt focussed on privacy management and linked to a perturbative method. Privacy must be considered not only as technical debt requirements but also at design and deployment phases, among others. In addition, this method is illustrated with a use case.

Autism Spectrum Disorder (ASD) is a largely prevalent neurodevelopmental condition with a big social and economical impact affecting the entire life of families. There is an intense search for biomarkers that can be assessed as early as possible in order to initiate treatment and preparation of the family to deal with the challenges imposed by the condition. Brain imaging biomarkers have special interest. Specifically, functional connectivity data extracted from resting state functional magnetic resonance imaging (rs-fMRI) should allow to detect brain connectivity alterations. Machine learning pipelines encompass the estimation of the functional connectivity matrix from brain parcellations, feature extraction, and building classification models for ASD prediction. The works reported in the literature are very heterogeneous from the computational and methodological point of view. In this paper, we carry out a comprehensive computational exploration of the impact of the choices involved while building these machine learning pipelines. Specifically, we consider six brain parcellation definitions, five methods for functional connectivity matrix construction, six feature extraction/selection approaches, and nine classifier building algorithms. We report the prediction performance sensitivity to each of these choices, as well as the best results that are comparable with the state of the art.

Abstract In the current context of an aging population in many developed countries, the issue of healthy aging is at the forefront of the political, scientific, and technological concerns. The frailty accompanying the late years of elderly people (>70 years old) deserves special consideration due to its great economical and personal costs and the workload imposed on the health care system. Hospital readmissions under a short time after hospital discharge are one of the sources of concern, and much effort is being devoted to their prediction for better care of the elder and optimized resource management. In this paper, we consider the prediction of readmissions for patients that are evaluated positively in the frailty scales. The computational experiments are carried out over a gender‐balanced cohort of 645 patients recruited at the University Hospital of Alava. We report machine‐learning prediction results of the readmission before the standard readmission limit of 30 days. We apply an upsampling technique to correct for class imbalance. Results are positive, encouraging further research and the creation of larger cohorts in international efforts.

Human activity recognition and neural activity analysis are the basis for human computational neureoethology research dealing with the simultaneous analysis of behavioral ethogram descriptions and neural activity measurements. Wireless electroencephalography (EEG) and wireless inertial measurement units (IMU) allow the realization of experimental data recording with improved ecological validity where the subjects can be carrying out natural activities while data recording is minimally invasive. Specifically, we aim to show that EEG and IMU data fusion allows improved human activity recognition in a natural setting. We have defined an experimental protocol composed of natural sitting, standing and walking activities, and we have recruited subjects in two sites: in-house ([Formula: see text]) and out-house ([Formula: see text]) populations with different demographics. Experimental protocol data capture was carried out with validated commercial systems. Classifier model training and validation were carried out with scikit-learn open source machine learning python package. EEG features consist of the amplitude of the standard EEG frequency bands. Inertial features were the instantaneous position of the body tracked points after a moving average smoothing to remove noise. We carry out three validation processes: a 10-fold cross-validation process per experimental protocol repetition, (b) the inference of the ethograms, and (c) the transfer learning from each experimental protocol repetition to the remaining repetitions. The in-house accuracy results were lower and much more variable than the out-house sessions results. In general, random forest was the best performing classifier model. Best cross-validation results, ethogram accuracy, and transfer learning were achieved from the fusion of EEG and IMUs data. Transfer learning behaved poorly compared to classification on the same protocol repetition, but it has accuracy still greater than 0.75 on average for the out-house data sessions. Transfer leaning accuracy among repetitions of the same subject was above 0.88 on average. Ethogram prediction accuracy was above 0.96 on average. Therefore, we conclude that wireless EEG and IMUs allow for the definition of natural experimental designs with high ecological validity toward human computational neuroethology research. The fusion of both EEG and IMUs signals improves activity and ethogram recognition.

Noninvasive behavior observation techniques allow more natural human behavior assessment experiments with higher ecological validity. We propose the use of gaze ethograms in the context of user interaction with a computer display to characterize the user's behavioral activity. A gaze ethogram is a time sequence of the screen regions the user is looking at. It can be used for the behavioral modeling of the user. Given a rough partition of the display space, we are able to extract gaze ethograms that allow discrimination of three common user behavioral activities: reading a text, viewing a video clip, and writing a text. A gaze tracking system is used to build the gaze ethogram. User behavioral activity is modeled by a classifier of gaze ethograms able to recognize the user activity after training. Conventional commercial gaze tracking for research in the neurosciences and psychology science are expensive and intrusive, sometimes impose wearing uncomfortable appliances. For the purposes of our behavioral research, we have developed an open source gaze tracking system that runs on conventional laptop computers using their low quality cameras. Some of the gaze tracking pipeline elements have been borrowed from the open source community. However, we have developed innovative solutions to some of the key issues that arise in the gaze tracker. Specifically, we have proposed texture-based eye features that are quite robust to low quality images. These features are the input for a classifier predicting the screen target area, the user is looking at. We report comparative results of several classifier architectures carried out in order to select the classifier to be used to extract the gaze ethograms for our behavioral research. We perform another classifier selection at the level of ethogram classification. Finally, we report encouraging results of user behavioral activity recognition experiments carried out over an inhouse dataset.

Grasping and object manipulation is a key element of intelligent behavior. Many innovative cyberphysical systems involve some kind of object grasp and manipulation, to the extent that grasping has been recognized as a critical technology for the next generation industrial systems. In this survey, we aim to draw a broad landscape of applications and current research trends and topics relating to grasping techniques and tools. Applications range from biomedical and surgical to industrial warehouse pick and place tasks, covering a wide range of spatial scales, from micro to macro scales. The resources involved and research lines under development include the latest computational intelligence tools as well as the research on new materials and devices for sensing and actuation.

Abstract The use of data driven predictive systems is becoming widespread as innovations in machine learning techniques have allowed the training of increasingly sophisticated models via the available data. The light detection and ranging (LiDAR) remote sensing technique is being increasingly applied to obtain informative terrain maps, due to its ability to collect large amounts of data with satisfactory accuracy. This paper focuses on the application of machine‐learning‐based predictive systems for the extraction of biomass information from LiDAR data. Biomass information has inmense ecological and economical value. We demonstrate the estimation of thePinus radiatabiomass in the Arratia‐Nervión region (Spain). Biomass estimation is considered a regression problem in which the ground truth for some specific sample sites is available. The promising results obtained in this study indicate that LiDAR data can be used to carry out detailed biomass mappings by the extrapolation of the models trained in this study.

Human motion capture by optical sensors produces snapshots of the motion of a cloud of points that need to be labeled in order to carry out ensuing motion analysis for medical or other purposes. We generate the labeling of instantaneous captures of the cloud of points, discarding temporal correlations, in the presence of occlusions. Our approach proposes an ensemble of weak classifiers defined over geometrical features extracted from small subsets of the cloud of points. We apply an Adaboost strategy to select a minimal ensemble of weak classifiers achieving a target correct labeling detection accuracy. Furthermore, we use these features to generate the labeling of the points in the cloud even in the presence of occlusions.To deal with the occlusions of markers we search for ensembles of partial labeling solvers which can provide partial consistent labelings which cover the unoccluded markers. We test two greedy search approaches and a genetic algorithm in the search for the optimal ensemble of partial solvers We demonstrate the approach on a real dataset obtained from the measurement of gait motion of persons, with available ground truth labeling. Results are encouraging, achieving high accuracy label generation at a reduced computational cost.

Emergency Departments (ED) suffer heavy overload due to lack of primary attention service. Increasingly geriatric admissions pose specific problems contributing to this overload. A consequence is the increase of patient returning short time after discharge, i.e., readmissions, sometimes requiring hospitalization. In this latter case the patient problem was not solved in the first admission and the condition has aggravated. The time threshold defining a patient comeback as readmission varies; therefore we have considered several such thresholds in our prediction experiments. Prediction of hospitalization following ED readmission is posed over a heavily imbalanced class distribution, so we have considered several approaches to deal with imbalanced datasets and several base classifiers, as well as performance measures that enhance the critical comparison between approaches. Experimental works are carried out on real data from a university hospital in Santiago, Chile, corresponding to a period of 3 years, including pediatric and adult admissions to the ED. We achieve results that encourage the development of real life application of the data balancing and classification approach for prediction of hospitalization after readmission.

Hallucinations are elusive phenomena that have been associated with psychotic behavior, but that have a high prevalence in healthy population. Some generative mechanisms of Auditory Hallucinations (AH) have been proposed in the literature, but so far empirical evidence is scarce. The most widely accepted generative mechanism hypothesis nowadays consists in the faulty workings of a network of brain areas including the emotional control, the audio and language processing, and the inhibition and self-attribution of the signals in the auditive cortex. In this paper, we consider two methods to analyze resting state fMRI (rs-fMRI) data, in order to measure effective connections between the brain regions involved in the AH generation process. These measures are the Dynamic Causal Modeling (DCM) cross-covariance function (CCF) coefficients, and the partially directed coherence (PDC) coefficients derived from Granger Causality (GC) analysis. Effective connectivity measures are treated as input classifier features to assess their significance by means of cross-validation classification accuracy results in a wrapper feature selection approach. Experimental results using Support Vector Machine (SVM) classifiers on an rs-fMRI dataset of schizophrenia patients with and without a history of AH confirm that the main regions identified in the AH generative mechanism hypothesis have significant effective connection values, under both DCM and PDC evaluation.

We present a cooperative aerial robot system for the transportation of hoses. The hose–robot attachment makes the whole system physically interconnected but not rigid, so that control design becomes a difficult nonlinear optimization problem. The hose in quasistationary state can be modeled by sections of catenary curves. We use proportional integral derivative (PID) controllers for both quadrotor attitude and trajectory control, tuned by particle swarm optimization (PSO). In this work we test PSO minimizing an energy function to achieve the PID controller tuning for horizontal motion of quadrotor teams transporting hoses under different stress conditions.

Electronic Health Records are clinical information repositories that have been proposed primarily to provide access to all clinical data of a patient. They have been formally defined by a dual model composed of a reference model and an archetype model. Such dual approach allows semantic interoperability, thus making different systems understand each other. In this work we extend the current structure with a third Decisional Model that will allow reasoning over the embedded clinical contents. Such reasoning will be based on the reuse of the clinical experience gained by the corresponding clinical professionals during different decision procedures.

Recent classification-oriented proposals to thematic maps building from hyperspectral images have used both semisupervised approaches and spatial information for correction of spectral classification. Semisupervised approaches enrich the training data set adding similar samples to each class, whereas spatial correction is based on the natural assumption of thematic class spatial compactness. In this paper, we propose and validate the following innovations: 1) a new spectral classifier, which is called anticipative hybrid extreme rotation forest (AHERF); 2) a spatial-spectral semisupervised approach; and 3) a final spatial classification correction step. The novel heterogeneous ensemble learning approach AHERF starts with a model selection phase, using a small subsample of the training data, in order to define a ranking-based selection probability distribution of the classifier architectures that will be used in the ensemble, so that the architecture best adapted to the data domain will be used more frequently to train individual classifiers in the ensemble. After this initial phase, AHERF trains a heterogeneous ensemble applying random rotations to bootstrapped samples of the remaining training data, aiming to obtain diversified and data-domain adapted individual classifiers. The natural assumption that spatially close pixels will most likely have highly correlated values is exploited in two phases of the process pipeline. First, semisupervised label assignment is supported by spectral similarity and spatial proximity. Unsupervised spectral similarity is detected by latent class discovery. In this paper, we use a clustering algorithm (i.e., k-means). Second, maximizing class spatial compactness removes classification errors that appear as speckle noise in the classification image. The whole approach aims to use minimal sets of labeled pixels for training, which we call the seed training data set. Testing results are computed over the entire image ground truth. For comparison, we provide results in several steps: 1) of classification by AHERF and competing classifiers built by semisupervised training and 2) after spatial correction. We validate the approach on several conventional benchmarking images, achieving results which are comparable with state-of-the-art approaches.

Multivariate mathematical morphology (MMM) aims to extend the mathematical morphology from gray scale images to images whose pixels are high-dimensional vectors, such as remote sensing hyperspectral images and functional magnetic resonance images (fMRIs). Defining an ordering over the multidimensional image data space is a fundamental issue MMM, to ensure that ensuing morphological operators and filters are mathematically consistent. Recent approaches use the outputs of two-class classifiers to build such reduced orderings. This paper presents the applications of MMM built on reduced supervised orderings based on lattice autoassociative memories (LAAMs) recall error measured by the Chebyshev distance. Foreground supervised orderings use one set of training data from a foreground class, whereas background/foreground supervised orderings use two training data sets, one for each relevant class. The first case study refers to the realization of the thematic segmentation of the hyperspectral images using spatial-spectral information. Spectral classification is enhanced by a spatial processing consisting in the spatial correction guided by a watershed segmentation computed by the LAAM-based morphological operators. The approach improves the state-of-the-art hyperspectral spatial-spectral thematic map building approaches. The second case study is the analysis of resting state fMRI data, working on a data set of healthy controls, schizophrenia patients with and without auditory hallucinations. We perform two experiments: 1) the localization of differences in brain functional networks on population-dependent templates and 2) the classification of subjects into each possible pair of cases. In this data set, we find that the LAAM-based morphological features improve over the conventional correlation-based graph measure features often employed in fMRI data classification.

The electronic health record (EHR) groups all digital documents related to a given patient as anamnesis, results of the laboratory tests, prescriptions, recorded medical signals as ECG or images etc. Dealing with such data representation we face with plethora of problems as different form of data, unstructured data (as doctor's notes), huge and fast growing volume, etc. It causes that EHR should be considered as the complex data representation. Accordingly, taking into consideration its complexity, hetorogenousity, fast growing and size we need special tools to analyse such medical big data. Such tools should be able to analyse datasets characterized by so-called 4Vs (volume, velocity, variety, and veracity). Notwithstandingly, we should also add the fifth V-value, because the only analytics tool deployment makes sense if it leads to healthcare improvement (as personalised patient's care, unnecessary hospitalization decreasing or reducing the patient's readmissions). In this paper we focus on the selected aspects EHR analysis from the big data perspective.

The Electronic Health Record (EHR) is becoming the central information object for various aspects of healthcare and medical related industries, from pharmaceuticals to bioengineering. This review provides a presentation of the state of affairs in several aspects of EHR, including security and privacy, data mining, design of decision support systems, acceptance by users and producers of health resources, and system implementation. In the last three years the number of publications has grown exponentially, therefore is rather difficult to be exhaustive, and the more technical aspects are expected to be quickly superseded by new advances.

Social network online services are growing at an exponential pace, both in quantity of users and diversity of services; thus, the evaluation of trust in the interaction among users and toward the system is a central issue from the user point of view. Trust can be grounded in past direct experience or in the indirect information provided by trusted third-party users shaping the trustee reputation. When there is no previous history of interactions, the truster must resort to some form of prediction in order to establish Trust or Distrust on a potential trustee. In this study, we deal with the prediction of trust relationships on the basis of reputation information. Trust can be positive or negative (Distrust), hence, we have a two-class problem. Feature vectors for the classification have binary-valued components. Artificial neural network and statistical classifiers provide state-of-the-art results with these features on a benchmarking trust database. In this article, we propose the application of a sample generation method for the minority class in order to reduce some of the effect of class imbalance among Trust and Distrust classes. Specifically, the approach shows high resiliency to system growth.

Decision recommendations are a set of alternative options for clinical decisions (e.g., diagnosis, prognosis, treatment selection, follow-up, and prevention) that are provided to decision makers by knowledge-based Clinical Decision Support Systems (k-CDSS) as aids. We propose to follow a “reasoning over domain” approach for the generation of decision recommendations by gathering and inferring conclusions from production rules. In order to rationalize our approach, we present a specification that will sustain the logic models supported in the knowledge bases we use for persistence. We introduce first the underlying knowledge model and then the necessary extensions that will convey toward the solution of the reported needs. The starting point of our approach is the proposition of Reflexive Ontologies (RO). Here, we go a step further, proposing an extension of RO that includes the handling and reasoning that production rules provide. Our approach speeds up the recommendation generation process.

Abstract Multiagent systems are increasingly present in computational environments. However, the problem of agent design or control is an open research field. Reinforcement learning approaches offer solutions that allow autonomous learning with minimal supervision. The Q‐learning algorithm is a model‐free reinforcement learning solution that has proven its usefulness in single‐agent domains; however, it suffers from dimensionality curse when applied to multiagent systems. In this article, we discuss two approaches, namely TRQ‐learning and distributed Q‐learning, that overcome the limitations of Q‐learning offering feasible solutions. We test these approaches in two separate domains. The first is the control of a hose by a team of robots. The second is the trash disposal problem. Computational results show the effectiveness of Q‐learning solutions to multiagent systems’ control.

Auditory hallucinations (AH) are a symptom that is most often associated with schizophrenia, but patients with other neuropsychiatric conditions, and even a small percentage of healthy individuals, may also experience AH. Elucidating the neural mechanisms underlying AH in schizophrenia may offer insight into the pathophysiology associated with AH more broadly across multiple neuropsychiatric disease conditions. In this paper, we address the problem of classifying schizophrenia patients with and without a history of AH, and healthy control (HC) subjects. To this end, we performed feature extraction from resting state functional magnetic resonance imaging (rsfMRI) data and applied machine learning classifiers, testing two kinds of neuroimaging features: (a) functional connectivity (FC) measures computed by lattice auto-associative memories (LAAM), and (b) local activity (LA) measures, including regional homogeneity (ReHo) and fractional amplitude of low frequency fluctuations (fALFF). We show that it is possible to perform classification within each pair of subject groups with high accuracy. Discrimination between patients with and without lifetime AH was highest, while discrimination between schizophrenia patients and HC participants was worst, suggesting that classification according to the symptom dimension of AH may be more valid than discrimination on the basis of traditional diagnostic categories. FC measures seeded in right Heschl's gyrus (RHG) consistently showed stronger discriminative power than those seeded in left Heschl's gyrus (LHG), a finding that appears to support AH models focusing on right hemisphere abnormalities. The cortical brain localizations derived from the features with strong classification performance are consistent with proposed AH models, and include left inferior frontal gyrus (IFG), parahippocampal gyri, the cingulate cortex, as well as several temporal and prefrontal cortical brain regions. Overall, the observed findings suggest that computational intelligence approaches can provide robust tools for uncovering subtleties in complex neuroimaging data, and have the potential to advance the search for more neuroscience-based criteria for classifying mental illness in psychiatry research.

This paper shows experimental results on electromyography (EMG)-based system control applied to motorized orthoses. Biceps and triceps EMG signals are captured through two biometrical sensors, which are then filtered and processed by an acquisition system. Finally an output/control signal is produced and sent to the actuators, which will then perform the actual movement, using algorithms based on autoregressive (AR) models and neural networks, among others. The research goal is to predict the desired movement of the lower arm through the analysis of EMG signals, so that the movement can be reproduced by an arm orthosis, powered by two linear actuators. In this experiment, best accuracy has achieved values up to 91%, using a fourth-order AR-model and 100ms block length.

In this article we present the design and implementation of a mobile cardiac monitoring system oriented to patients in Phase II and III of cardiac rehabilitation. The complete monitoring system involves both hardware and software design perspectives. At the hardware level, we present a T-shirt with a 12-lead ECG system and an embedded inertial sensor for the monitoring of activity and energy expenditure. At the software level, a modular cloud platform performs data processing to detect relevant cardiac events and to provide advanced visualization capabilities. As a case study, we have implemented our system at the Cardiac Rehabilitation program at Donostia University Hospital (Spain). Finally, the validation of the 12-lead ECG recording system is also presented and discussed.

At a time when socialism as an economic option is variously questioned, very few people are against social instances of our life such as entertainment, customer assistance, and so on. This happens with the management of many things accompanying our life as well. We can find both the reason and the evidence for the viability of this trend in one very basic fact: things are social because they work better. However, also in this sphere social politics are highly questionable. Here we introduce the perspective adopted in the European project SandS within a framework of Internet of Things. In this case things are agents interacting on the network within a service centric approach where a sound hierarchy dispatches instructions. It is a complete ecosystem where the social network develops a collective intelligence subtending new concrete functionalities that are centered on the user willing and fostered by his/her feedbacks. The central role of the user reflects on all aspects of the ecosystem, from the family of things which are socially governed: the household appliances (the white goods) that affect our everyday life, up to the employed hardware and software: strictly open source.

Ontology processing is arguably a time-consuming process with high associated computational costs. Query actions constitute a crucial part of the reasoning process and are a primary source of time consumption. Reflexive ontologies (ROs) is a novel approach intended to reduce time consumption problems while providing a fast reaction from ontology-based applications. In this article we present the implementation of a knowledge-based clinical decision support system (CDSS) for the diagnosis of Alzheimer's disease, which was the benchmark used to evaluate the impact of RO in the overall performance of the system. The implementation details and the definition of the implementation methodology are exposed in this article, along with the results of the evaluation. Some novel techniques that aim to optimize the performance of ROs are also presented with highlights of the test application introduced in our previous work.

Image interest point extraction and matching across images is a commonplace task in computer vision–based applications, across widely diverse domains, such as 3D reconstruction, augmented reality, or tracking. We present an empirical evaluation of state-of-the-art interest point detection algorithms measuring several parameters, such as efficiency, robustness to image domain geometric transformations—that is, similarity—affine or projective transformations, as well as invariance to photometric transformations such as light intensity or image noise.

This article uses an innovative approach for the identification of groups of users in a social network by solving a graph coloring problem (GCP). We focus on social networks of users of a company's intranet. We want to identify user groups sharing the same behavior while interacting with the enterprise resource planning (ERP) as a means to verify that their roles, work practices, and positions held in the company are correct for each user. From the ERP logs, we generate the internal social network (ISN) graph where each vertex corresponds to a user and edges correspond to relations between users. Two users will be related if they frequently use the same programs. Other user attributes like the role assigned by the administrator, workplace location, or department offer secondary information to create the relations between users. After generating the graph we identify the users' groups solving a GCP on the ISN complementary graph. Moreover, ISN graph topology changes over time, so the composition of the groups of users may also change. New users can enter the intranet and exit it, and the behavior of the users may change. Therefore, we need approaches the quickly provide user clustering solutions to be able to adapt to the ISN time changes. We propose the gravitational swarm for graph coloring (GSGC) for this task and compare it with state-of-the-art approaches. We provide results on real-life (anonymized) cases.

New approaches are needed that could move us toward developing effective systems for problem solving and decision making, systems that can deal with complex and ill-structured situations, systems t...

This Special Edition follows Volume I (Current Research Advances and Implementations in Smart Knowledge-Based Systems: Part I) and contains carefully selected and reviewed papers that expand signif...

This paper proposes a Hybrid Extreme Rotation Forest (HERF) classifier for segmentation of 3D Computed Tomography Angiography (CTA) following an Active Learning (AL) approach. The HERF is an ensemble of classifiers composed of Extreme Learning Machines (ELM) and Decision Trees. Training of the HERF includes optimal rotation of random partitions of the feature set aimed to increase diversity. AL follows an strategy of optimal sample selection in order to minimize the number of data samples needed to obtain a required accuracy degree. AL is pertinent for interactive learning processes where a human operator is required to select training samples to enhance the classifier in an iterative process, therefore labeling samples for training may be a time consuming and expensive process. CTA image segmentation is one of such processes, due to the variability in CTA images which hinders the generalization of classifiers trained on one dataset to new datasets. Following an AL strategy, the human operator is presented with a visual selection of pixels whose labeling would be most informative for the classifier. After adding those labeled pixels to the training data, the classifier is retrained. This iteration is repeated until image segmentation quality meets the required level. The approach is applied to the segmentation of the thrombus in CTA data of Abdominal Aortic Aneurysm (AAA) patients, showing that the structures of interest in CTA volume can be accurately segmented after a few iterations using a small data sample.

Lattice Computing is the class of algorithms built on the basis of Lattice Theory. They either perform operations in the ring of the real valued spaces endowed with some (inf, sup) lattice operators, or use lattice theory to produce generalizations or fusions of conventional approaches. Lattice Computing has produced a variety of algorithms for data processing, classification, signal filtering over the last decades. On the other hand, hybrid algorithms are flourishing in the last years giving innovative solutions to new and old problems. Hybrid algorithms are free combinations of Computational Intelligence approaches for data mining, signal processing or general artificial intelligence questions, including statistical, nature and bio-inspired algorithms. In this paper we review some Lattice Computing approaches and how they have been hybridized for specific problems.

Analysis of fMRI data, specifically resting-state fMRI data, is performed here from the point of view of a hybrid Multivariate Mathematical Morphology induced by a supervised h-ordering defined on the fMRI time series by the response of Lattice Auto-associative Memories built from specific fMRI voxels. The supervised h-ordering values and the results of morphological filters, i.e. a morphological top-hat, allow to identify some brain networks depending on the seed voxel value. Results on a set of resting state fMRI images of schizophrenia patients and healthy controls show that these networks can be dependent on the subject class, thus providing discriminant findings that may be useful for machine learning approaches.

Abdominal Aortic Aneurysm (AAA) is a local dilation of the Aorta that occurs between the renal and iliac arteries. The weakening of the aortic wall leads to its deformation and the generation of a thrombus. Recently developed treatment involves the insertion of a endovascular prosthetic (EVAR), which has the advantage of being a minimally invasive procedure but also requires monitoring to analyze postoperative patient outcomes using 3D Contrast Computerized Tomography Angiography (CTA) imaging procedures. In order to effectively assess the changes experienced after surgery, it is necessary to segment the aneurysm in the CT volume, which is a very time-consuming task. Here we provide results of a novel active learning approach for the semi-automatic detection and segmentation of the lumen and the thrombus of the AAA, which uses image intensity features and discriminative Random Forest classifiers.

The karyotyping step is essential in the genetic diagnosis process, since it allows the genetician to see and interpret patient’s chromosomes. Today, this step of karyotyping is a time-cost procedure, especially the part that consists in segmenting and classifying the chromosomes by pairs. This paper presents a compartive study of image classification of banded human chromosomes for automated karyotyping (AKS), by using classifier ensembles. The goal of this contribution is to propose and evaluate a solution to automate the karyotyping, from microscope images to the obtention of the classified chromosomes. For this purpose, we have evaluated several approaches based on classifier ensembles trying to find a solution that shows better trade-off between accuracy and computational cost.

We give a review of recent literature on the subject of trust in computational interactions between human and artificial autonomous agents, or between artificial autonomous agents. Trust is a central question in the development of an ethical approach to the design of multiagent autonomous systems, which must lead to safe and sound interactions, and the development of secure systems based on ethically sound behavior design. The topic treatment in the literature is extensive and from a multitude of points of view, therefore the presented review may be falling short in many regards. We have focused on two main axes to organize the references: computational models and applications.

Clinical Decision Support Systems (CDSS) are computer applications that focus on assisting medical decisions required during clinical tasks. Although CDSS have been extensively studied for more than 30 years, their use is not broadly extended yet in daily clinical practice. Identified challenges of CDSS include (i) automating decision support, (ii) clinical workflow integration, (iii) ability of the system to be maintained and extended, (iv) timely advice and (v) evaluation of decisions effects and costs. In this paper, we hypothesize that Knowledge Engineering techniques and semantic technologies could be applied to CDSS in order to overcome the current identified challenges. We present a generic architecture for a Semantic CDSS, which we call SCDSS, and implementation guidelines for the breast cancer domain. Our approach follows a cyclic-federated paradigm allowing the reutilization of knowledge gathered at every stage of the clinical cycle.

Single robot hose transport is a limit case of linked multicomponent robotic systems, where one robot moves the tip of a hose to a desired position. The interaction between the passive, flexible hose and the robot introduces highly nonlinear effects in the system's dynamics, requiring innovative control design approaches, such as reinforcement learning. This article improves previous approaches to this problem by introducing a novel reinforcement learning algorithm (TRQ-learning) and a new system state definition for the autonomous derivation of the hose–robot control algorithm. Computational experiments based on accurate geometrically exact dynamic splines hose dynamics simulations show the improvement obtained.

In this article we present an experience-based clinical decision support system (CDSS) for the diagnosis of Alzheimer's disease, which enables the discovery of new knowledge in the system and the generation of new rules that drive reasoning. In order to evolve an initial set of production rules given by medical experts we make use of the Set of Experience Knowledge Structure (SOEKS). An illustrative case of our system is also presented.

The use of engineering standards in virtual engineering and their potential as models for the specification of a given domain's ontology are arguably unexplored. The importance of domain modeling in virtual engineering deals directly with the potential benefits that the semantic technologies may bring, allowing to discover implicit knowledge that can be beneficial for engineers. This work presents a state-of-the-art review of the technologies used in our approach, a successful case study where our methodology was applied, and the description and results of an experiment designed to provide a quantitative validation of our methodology.

Virtual engineering applications (VEA) are used through the whole product life cycle (PLC) process; their complexity and pervasiveness make them an ideal scenario for development and testing of software engineering innovations. More precisely, VEAs have problems when they need to share their specific knowledge. They suffer semantic loss situations, and they hardly use any semantic tools. This work proposes a new approach to solving these kinds of problems based on semantic technologies, allowing the seamless sharing of information and knowledge between VEAs involved in a PLC scenario. This approach is validated through a plant layout design application, where several VEA can cooperate and share their knowledge, accomplishing the design task.

We present a color gradient with good color constancy preservation properties. The approach does not need a priori information or changes in color space. It is based on the angular distance between pixel color representations in the RGB space. It is naturally invariant to intensity magnitude, implying high robustness against bright spots produced be specular reflections and dark regions of low intensity.

We discuss some aspects of a well known algorithm for inhomogeneity intensity correction in Magnetic Resonance Imaging (MRI), the parametric bias correction (PABIC) algorithm. In this approach, the intensity inhomogeneity is modelled by a linear combination of 2D or 3D Legengre polynomials (computed as outer products of 1D polynomials). The model parameter estimation process proposed in the original paper is similar to a (1+1) Evolution Strategy, with some small and subtle differences. In this paper we discuss some features of the algorithm elements, trying to uncover sources of undesired behaviors and the limits to its applicability. We study the energy function proposed in the original paper and its relation to the image formation model. We also discuss the original minimization algorithm behavior. We think that this detailed discussion is needed because of the high impact that the original paper had in the literature, leading to an implementation into the well known ITK library, which means that it has become a de facto standard.

Defining lattice computing as the class of algorithms that either apply lattice operators inf and sup or use lattice theory to produce generalizations or fusions of previous approaches, we find that a host of algorithms for data processing, classification, signal filtering, have been produced over the last decades. We give a fast and brief review, which by no means could be exhaustive; with the aim of showing that this area has been growing during the past decades and to highlight the ones that we think are broad avenues for future research. Although our emphasis is on Artificial Neural Networks and Fuzzy Systems in this review we include Mathematical Morphology as a notorious instance of Lattice Computing.

Lattice associative memories have been proposed for image denoising and pattern recognition. We have shown that they can be applied to other domains, like image retrieval and hyperspectral image unsupervised segmentation. In both cases the key idea is that autoassociative morphological memories selective sensitivity to erosive and dilative noise can be applied to detect the lattice independence between patterns. The convex coordinates obtained by linear unmixing based on the sets of lattice independent patterns define a feature extraction process. These features may be useful either for content based image retrieval (CBIR) as well as for pattern classification. We present some CBIR and recognition results on a mushroom shape database, including the comparison with other linear feature extraction algorithms (ICA and CCA).

In recent years, the concern with image authentication and ownership issues is growing in the remote sensing community. Watermarking techniques help to solve the problems raised by this issue. In this paper we elaborate on the proposition of an optimal placement of the watermark image in a hyperspectral image. The problem is posed as a multi-objective optimization problem. Two objective functions are formulated, one modeling the robustness of the watermark recovery and other modeling the distortion introduced in the watermarked image. The application of an evolutionary algorithm (MOGA) to the optimal watermarking hyperspectral images is presented.

Convergence of the self-organizing map (SOM) and neural gas (NG) is usually contemplated from the point of view of stochastic gradient descent (SGD) algorithms of an energy function. SGD algorithms are characterized by very restrictive conditions that produce a slow convergence rate. Also they are local minimization algorithms, very dependent on the initial conditions. However, some empirical results show that one-pass on-line training realizations of SOM and NG may perform comparable to more careful (slow) realizations. Moreover, other empirical works suggest that SOM is quite robust against initial conditions. In both cases the performance measure is the quantization distortion. That empirical evidence leads us to propose that the appropriate setting for the convergence analysis of SOM NG and similar competitive artificial neural network clustering algorithms is the theory of graduated non-convexity (GNC) algorithms.

We propose an evolutionary algorithm for the enhancement of digital semi-fragile watermaking based on the manipulation of the image discrete cosine transform (DCT). The algorithm searches for the optimal localization of the DCT of an image to place the mark image DCT coefficients. The problem is stated as a multi-objective optimization problem (MOP), that involves the simultaneous minimization of distortion and robustness criteria.

Memetic algorithms are hybrid evolutionary algorithms that combine local optimization with evolutionary search operators. In this paper we describe an instance of this paradigm designed for the correction of illumination inhomogeneities in images. The algorithm uses the gradient information of an error function embedded in the mutation operator. Moreover, the algorithm is a single-solution population algorithm, which makes it computationally light. The fitness function is defined assuming that the image intensity is piecewise constant and that the illumination bias may be approximated by a linear combination of 2D Legendre polynomials. We call the algorithm instantaneous memetic illumination correction (IMIC).

Autoassociative morphological memories (AMM) are a construct similar to hopfield autoassociatived memories defined on the (R, +, v, /spl and/) lattice algebra. Unlimited storage and perfect recall of noiseless real valued patterns has been proved for AMMs. However AMMs suffer from sensitivity to specific noise models, that can be characterized as erosive and dilative noise. On the other hand, spectral unmixing of hyperspectral images needs the prior definition of a set of endmembers, which correspond to material spectra lying on vertices of the minimum convex region covering the image data. These vertices can be characterized as morphologically independent patterns. We present a procedure based on the AMM noise sensitivity for endmember detection based on this characterization.

We propose a procedure for extraction of spectra from hyperspectral images that may be used as endmembers for unmixing which uses the autoassociative morphological memories (AMM) as detectors of morphological independence conditions. Endmember spectra correspond to vertices of a convex region that covers the image pixel spectra. The morphological independence, after shifting the data to zero mean, is a necessary condition for these vertices. The selective sensitivity of AMM to noise characterized as erosive and dilative noise allows their use as morphological independence detectors.

Spectral unmixing of hyperspectral images relies on the knowledge of a set of endmembers, which are usually unknown. One approach is the induction from the image data of the endmember spectra. Endmember spectra correspond to vertices of a convex region that covers the image pixel spectra. The morphological independence, after shifting the data to zero mean, is a necessary condition for these vertices. We propose a procedure for extraction of spectra from hyperspectral images that may be used as endmembers for unmixing which uses the Autoassociative Morphological Memories (AMM) as detectors of morphological independence conditions.

Morphological neural networks (MNN) have been proposed as an alternative neural computation paradigm. We explore the potential of heteroassociative MNN (HMNN) for a practical task, such as that of robust scene recognition. Scene recognition could be of use for self-localization in a vision-based navigation framework for mobile robots. HMNN have a big potential for real time application because its recall process is very fast. We present some experimental results that illustrate our ideas.

Morphological neural networks (MNN) have been proposed as associative memories (with its two cases: autoassociative and heteroassociative). In this paper we are involved with heteroassociative MNN (HMNN). We propose their use for self-localization in a vision-based navigation framework for mobile robots. HMNN can be trained in a single computation step. Their storage capacity bound is the dimension of the patterns, and they have perfect recall of the patterns under very mild conditions. Recall is also very fast, because the MNN recall does not involve the search for an energy minimum.

In this paper we propose the application of vector quantizers computed over an image for its own filtering. Each pixel processing is conditioned to its neighborhood and the neighborhood's closest code vector. The code vectors play the role of conditional context in a Bayesian image processing framework. The approach is applied to high resolution MRI. The visual results show that this approach produces image smoothing with good edge preservation, although no edge model is introduced.

The hit-or-miss transform (HMT) is a powerful morphological tool for the processing of binary images. There have been several attempts to generalize it to grayscale images, based on the grayscale erosion. The goal is to obtain a translation invariant recognition tool, with some robustness regarding small deformations and variations of illumination. We propose a definition of the hit-or-miss transform based on level sets. We call it the level set hit-or-miss transform (LSHMT). We compare its performance with that of another grayscale HMT found in the literature. The task performed is that of face localization on grayscale images, based on a set of face patterns. The empirical results on a database show the greater robustness of the LSHMT. The generalization of LSHMT using connected and morphological operators by reconstruction are proposed as feasible lines of research to increase its robustness. We are also working in its generalization to color images and image sequences.

We propose the use of morphological multi-scale fingerprints (MMF) for face localization. The MMF is computed as the local maxima and minima preserved up to a certain scale in a multi-scale analysis based on morphological erosion and dilation. This approach belongs to a class of global image feature extraction approaches, that can be combined with others to ensure robust face localization. No structural relationships between face elements is taken into account. We compare this approach to the eigenface approach to face detection with clear superior results.

To increase the robustness of visual processing in the context of mobile robotics, we introduce an image filtering process based on the codebooks computed by the SOM. The Self Organizing Map and the Simple Competitive Learning are used to compute adaptively the vector quantizers of color image sequences. The codebook computed for each image in the sequence is then used as a smoothing filter, the VQ Bayesian Filter (VQ-BF), for the preprocessing of the images in the sequence. This filter is applied to the computation of optical flow at the single pixel level.

Morphological Neural Networks (MNN) have been proposed as associative (with its two cases: autoassociative and heteroassociative) memories. In this paper we are involved with Heteroassociative MNN (HMNN). We propose their utilization as a preprocessing step for human shape detection, in a vision-based navigation problem for mobile robots. MNN can be trained in a single computing step, they possess unlimited storing capacity, and they have perfect recall of the pattens. Recall is also very fast, because the MNN recall does not involve the search for an energy minimum.

In this paper we consider the determination of the structure of the high-order Boltzmann machine (HOBM), a stochastic recurrent network for approximating probability distributions. We obtain the structure of the HOBM, the hypergraph of connections, from conditional independences of the probability distribution to model. We assume that an expert provides these conditional independences and from them we build independence maps, Markov and Bayesian networks, which represent conditional independences through undirected graphs and directed acyclic graphs respectively. From these independence maps we construct the HOBM hypergraph. The central aim of this paper is to obtain a minimal hypergraph. Given that different orderings of the variables provide in general different Bayesian networks, we define their intersection hypergraph. We prove that the intersection hypergraph of all the Bayesian networks (N!) of the distribution is contained by the hypergraph of the Markov network, it is more simple, and we give a procedure to determine a subset of the Bayesian networks that verifies this property. We also prove that the Markov network graph establishes a minimum connectivity for the hypergraphs from Bayesian networks.

In this paper we give a formal definition of the high-order Boltzmann machine (BM), and extend the well-known results on the convergence of the learning algorithm of the two-order BM. From the Bahadur-Lazarsfeld expansion we characterize the probability distribution learned by the high order BM. Likewise a criterion is given to establish the topology of the BM depending on the significant correlations of the particular probability distribution to be learned.

Generalized learning vector quantization (GLVQ) has been proposed in as a generalization of the simple competitive learning (SCL) algorithm. The main argument of GLVQ proposal is its superior insensitivity to the initial values of the weights (code vectors). In this paper we show that the distinctive characteristics of the definition of GLVQ disappear outside a small domain of applications. GLVQ becomes identical to SCL when either the number of code vectors grows or the size of the input space is large. Besides that, the behavior of GLVQ is inconsistent for problems defined on very small scale input spaces. The adaptation rules fluctuate between performing descent and ascent searches on the gradient of the distortion function.

Boltzmann machines (BMs) are proposed as a computational model for the solution of the satisfiability (SAT) problem in the propositional calculus setting. Conditions that guarantee consensus function maxima for configurations of the BM associated with solutions to the satisfaction problem are given. Experimental results that show a linear behavior of BMs solving the satisfiability problem are presented and discussed.>