Search

The file attached to this record is the author's final peer reviewed version. The Publisher's final version can be found by following the DOI link. In the context of the global coronavirus pandemic, different deep learning solutions for infected subject detection using chest X-ray images have been proposed. However, deep learning models usually need large labelled datasets to be effective. Semi-supervised deep learning is an attractive alternative, where unlabelled data is leveraged to improve the overall model’s accuracy. However, in real-world usage settings, an unlabelled dataset might present a different distribution than the labelled dataset (i.e. the labelled dataset was sampled from a target clinic and the unlabelled dataset from a source clinic). This results in a distribution mismatch between the unlabelled and labelled datasets. In this work, we assess the impact of the distribution mismatch between the labelled and the unlabelled datasets, for a semi-supervised model trained with chest X-ray images, for COVID-19 detection. Under strong distribution mismatch conditions, we found an accuracy hit of almost 30%, suggesting that the unlabelled dataset distribution has a strong influence in the behaviour of the model. Therefore, we propose a straightforward approach to diminish the impact of such distribution mismatch. Our proposed method uses a density approximation of the feature space. It is built upon the target dataset to filter out the observations in the source unlabelled dataset that might harm the accuracy of the semi-supervised model. It assumes that a small labelled source dataset is available together with a larger source unlabelled dataset. Our proposed method does not require any model training, it is simple and computationally cheap. We compare our proposed method against two popular state of the art out-of-distribution data detectors, which are also cheap and simple to implement. In our tests, our method yielded accuracy gains of up to 32%, when compared to the previous state of the art methods. The good results yielded by our method leads us to argue in favour for a more data-centric approach to improve model’s accuracy. Furthermore, the developed method can be used to measure data effectiveness for semi-supervised deep learning model training.

Skin lesions are caused due to multiple factors, like allergies, infections, exposition to the sun, etc. These skin diseases have become a challenge in medical diagnosis due to visual similarities, where image classification is an essential task to achieve an adequate diagnostic of different lesions. Melanoma is one of the best-known types of skin lesions due to the vast majority of skin cancer deaths. In this work, we propose an ensemble of improved convolutional neural networks combined with a test-time regularly spaced shifting technique for skin lesion classification. The shifting technique builds several versions of the test input image, which are shifted by displacement vectors that lie on a regular lattice in the plane of possible shifts. These shifted versions of the test image are subsequently passed on to each of the classifiers of an ensemble. Finally, all the outputs from the classifiers are combined to yield the final result. Experiment results show a significant improvement on the well-known HAM10000 dataset in terms of accuracy and F-score. In particular, it is demonstrated that our combination of ensembles with test-time regularly spaced shifting yields better performance than any of the two methods when applied alone. Yes

In this work we implement a COVID-19 infection detection system based on chest Xray images with uncertainty estimation. Uncertainty estimation is vital for safe usage of computer aided diagnosis tools in medical applications. Model estimations with high uncertainty should be carefully analyzed by a trained radiologist. We aim to improve uncertainty estimations using unlabelled data through the MixMatch semi-supervised framework. We test popular uncertainty estimation approaches, comprising Softmax scores, Monte-Carlo dropout and deterministic uncertainty quantification. To compare the reliability of the uncertainty estimates, we propose the usage of the Jensen-Shannon distance between the uncertainty distributions of correct and incorrect estimations. This metric is statistically relevant, unlike most previously used metrics, which often ignore the distribution of the uncertainty estimations. Our test results show a significant improvement in uncertainty estimates when using unlabelled data. The best results are obtained with the use of the Monte Carlo dropout method.

Foreground object detection is a fundamental low level task in current video surveillance systems. It is usually accomplished by keeping a model of the background at each frame pixel. Many background learning algorithms have difficulties to attain real time operation when applied directly to the output of state of the art high resolution surveillance cameras, due to the large number of pixels. Here we propose a strategy to address this problem which consists in maintaining a low resolution model of the background which is upscaled by adaptive super resolution in order to produce a foreground detection mask of the same size as the original input frame. Extensive experimental results demonstrate the suitability of our proposal, in terms of reduction of the computational load and foreground detection accuracy. Universidad de Málaga. Campus de Excelencia Internacional Andalucía Tech.

Most classic approaches to homography estimation are based on the filtering of outliers by means of the RANSAC method. New proposals include deep convolutional neural networks. Here a new method for homography estimation is presented, which supplies a deep neural homography estimator with color perturbated versions of the original image pair. The obtained outputs are combined in order to obtain a more robust estimation of the underlying homography. Experimental results are shown, which demonstrate the adequate performance of our approach, both in quantitative and qualitative terms. Universidad de Málaga. Campus de Excelencia Internacional Andalucía Tech.

CCI Group has contributed to the research Traffic flow related air pollution is one of the major problems in urban areas, and is often difficult to avoid it if the time sequenced dynamic pollution and traffic parameters are not identified and modelled efficiently. In our introduced work here, an artificial intelligence technique such as Bayesian networks are used for a robust traffic data analysis and modelling. The most common challenge in traditional data analysis is a lack of capability of unveiling the hidden links between the distant data attributes (e.g. pollution sources, dynamic traffic parameters, geographic location characteristics, etc.), whereas some subtle effects of these parameters or events may play an important role in pollution on a long term basis.

The supervised descriptive rule discovery concept groups a set of data mining techniques whose objective is to describe data with respect to a property of interest. Among the techniques within this concept are the subgroup discovery, emerging patterns and contrast sets. This contribution presents the supervised descriptive rule discovery concept within the data mining literature. Specifically, it is important to remark the main di erence with respect to other existing techniques within classification or description. In addition, a a survey of the state-of-the-art about the different techniques within supervised descriptive rule discovery throughout the literature can be observed. The paper allows to the experts to analyse the compatibilities between terms and heuristics of the different data mining tasks within this concept.

El siguiente es un estudio descriptivo el cual presenta la tendencia de cinco estudiantes nativo-hablantes del criollo limonense de usar fonemas del criollo cuando tratan de hablar en inglés en la escuela. Este estudio se llevó a cabo en el IEGB Limón 2000 en Liverpool, Limón durante los meses de Setiembre y Octubre del 2015 con la participación de la clase del 5-1. Además, la opinión de los padres, las maestras de inglés, la directora, y también del Asesor Regional de Ingles en Limón y de un sociólogo sobre el reconocimiento del criollo limonense, como parte de la cultura limonense ayudó a obtener una visión contextualizada de esta investigación. Para recoger la información necesaria, se utilizaron diferentes métodos de recolección de datos tales como observaciones, cuestionarios, entrevistas, y material audio-visual. Los resultados mostraron que si existe interferencia fonológica del criollo limonense en la producción de inglés en el aula; sin embargo, esta interferencia no afecta negativamente la compresibilidad del mensaje en el proceso de comunicación. Con respecto a la importancia de la enseñanza del criollo limonense en las aulas de las escuelas en Limón como una forma de revivir la cultura criolla en esa provincia, las respuestas de los participantes fueron diferentes sin importar el origen étnico de los participantes. The following is a descriptive study which presents the tendency of five Limonese Creole native speaking students to use Limonese Creole phonemes when trying to produce in English. This study took place at IEGB Limón 2000 Elementary School in Liverpool, Limon during the months of September and October of 2015 with the participation of the English teacher and students from the 5-1 grade. Moreover, the opinion of their parents, English teachers, and the Principal from that school, and also from the English Regional Advisor in Limon and a sociologist towards acknowledging Limonese Creole as part of the Limonese Culture attempted to obtain a contextualized vision of this research. In order to gather the necessary information, different methods of collecting data were used, such as observations, questionnaires, interviews, and flash cards. The results showed that there is language transfer in the phonological domain of Limonese Creole to Standard English as taught in the classrrom; the use of phonemes from Limonese Creole, however, does not impede or break the communication attemped in the target language. Regarding the importance of Limonese Creole in the Limon province and its incorporation as a formal subject as a means to revitalize the Creole culture in Limon, the impressions from the participants was not homogeneous, nevertheless the ethnic origin. Escuela de Literatura y Ciencias del Lenguaje

In this work a research is presented aimed to improve the mechanical performance models used to establish a range-extension methodology, and to introduce the use of computational intelligence to operate a real-time range extension engine management system to replace the current algorithmic approach. This paper describes the initial stage in design of the control strategy, taking into account a number of environmental factors in order to increase the range of series hybrid electric vehicles.

This paper presents experiments in the use of an Electromyographic sensor to determine whether a person is standing, walking or running. The output of the sensor was captured and processed in a variety of different ways to extract those features that were seen to be changing as the movement state of the person changed. Experiments were carried out by adjusting the parameters used for the collection of the features. These extracted features where then passed to a set of Artificial Neural Networks trained to recognise each state. This methodology exhibits an accuracy needed to control a prosthetic leg.

Computational Intelligence (CI) techniques are strong and efficient and capable of enhancing current traffic and air quality management systems.

Nowadays, there is a increasing in the use of learning management systems from the universities. This type of systems are also known under other di erent terms as course management systems or learning content management systems. Speci cally, these systems are e-learning platforms o ering di erent facilities for information sharing and communication between the participants in the e-learning process. This contribution presents an experimental study with several subgroup discovery algorithms based on evolutionary fuzzy systems using data from a web-based education system. The main objective of this contribution is to extract unusual subgroups to describe possible relationships between the use of the e-learning platform and marks obtained by the students. The results obtained by the best performing algorithm, NMEEF-SD, are also presented. The most representative results obtained by this algorithm are summarised in order to obtain knowledge that can allow teachers to take actions to improve student performance.

There is a strong need for local authorities to meet the challenges of sustainable transport as well as complying with air quality targets. iTRAQ - Integrated Traffic Management and Air Quality using Space Services is a European Space Agency funded project. During the course of a feasibility study a consortium of UK industry, academic and local authority partners developed and validated a dynamic system for optimising the use of the road network balanced with the need to sustain high standards of air quality. iTRAQ uses a number of inputs that enable it to sense the current situation in near-real-time and provide accurate forecasts using a computational intelligence module. Traffic flow, queues, and congestion are gathered using traditional ground-based sensors as well as Global Navigation Satellite Systems based vehicle data. Air quality information is obtained from in situ monitors, a City-wide Gaussian dispersion model (Airviro), a European-scale ensemble model (MACC), and direct measurements from low-earth orbit satellites (OMI and GOME-2). The concept has been tested and validated using near-real-time data and a simulation environment, providing enhanced strategies to the local authority. Following a successful feasibility study, a larger demonstration phase is now being planned and other local authorities are being encouraged to participate. This paper gives an overview of this novel system and presents some initial test results that confirm the feasibility of this integrated system, reducing the traffic delay, increasing the flow and optimising the local air quality levels.