The subject of this thesis is to develop a temporal neuro-fuzzy system for fore- casting the future behavior of a multivariate time series data. The system has two components combined by means of a system interface. First, a rule extraction method is designed which is named Fuzzy MAR (Multivari- ate Auto-regression). The method produces the temporal relationships between each of the variables and past values of all variables in the multivariate time series system in the form of fuzzy rules. These rules may constitute the rule-base in a fuzzy expert system. Second, a temporal neuro-fuzzy system which is named ANFIS unfolded in - time is designed in order to make the use of fuzzy rules, to provide an environment that keeps temporal relationships between the variables and to forecast the future behavior of data. The rule base of ANFIS unfolded in time contains temporal TSK(Takagi-Sugeno-Kang) fuzzy rules. In the training phase, Back-propagation learning algorithm is used. The system takes the multivariate data and the num- ber of lags needed which are the output of Fuzzy MAR in order to describe a variable and predicts the future behavior. Computer simulations are performed by using synthetic and real multivariate data and a benchmark problem (Gas Furnace Data) used in comparing neuro- fuzzy systems. The tests are performed in order to show how the system efficiently model and forecast the multivariate temporal data. Experimental results show that the proposed model achieves online learning and prediction on temporal data. The results are compared by other neuro-fuzzy systems, specifically ANFIS.

Predicting the next location of a user has been interesting for both academia and industry. Applications like location-based advertising, traffic planning, intelligent resource allocation as well as in recommendation services are some of the problems that many are interested in solving. Along with the technological advancement and the widespread usage of electronic devices, many location-based records are created. Today, deep learning framework has successfully surpassed many conventional methods in many learning tasks, most known in the areas of image and voice recognition. One of the neural network architecture that has shown the promising result at sequential data is Recurrent Neural Network (RNN). Since the creation of RNN, much alternative architecture have been proposed, and architectures like Long Short Term Memory (LSTM) and Gated Recurrent Units (GRU) are one of the popular ones that are created[5]. This thesis uses GRU architecture and features that incorporate time and location into the network to forecast people’s next location In this paper, a spatial-temporal neural network (ST-GRU) has been proposed. It can be seen as two parts, which are ST and GRU. The first part is a feature extraction algorithm that pulls out the information from a trajectory into location sequences. That process transforms the trajectory into a friendly sequence format in order to feed into the model. The second part, GRU is proposed to predict the next location given a user’s trajectory. The study shows that the proposed model ST-GRU has the best results comparing the baseline models. / Att förutspå vart en individ är på väg har varit intressant för både akademin och industrin. Tillämpningar såsom platsbaserad annonsering, trafikplanering, intelligent resursallokering samt rekommendationstjänster är några av de problem som många är intresserade av att lösa. Tillsammans med den tekniska utvecklingen och den omfattande användningen av elektroniska enheter har många platsbaserade data skapats. Idag har tekniken djupinlärning framgångsrikt överträffat många konventionella metoder i inlärningsuppgifter, bland annat inom områdena bild och röstigenkänning. En neural nätverksarkitektur som har visat lovande resultat med sekventiella data kallas återkommande neurala nätverk (RNN). Sedan skapandet av RNN har många alternativa arkitekturer skapats, bland de mest kända är Long Short Term Memory (LSTM) och Gated Recurrent Units (GRU). Den här studien använder en modifierad GRU där man bland annat lägger till attribut såsom tid och distans i nätverket för att prognostisera nästa plats. I det här examensarbetet har ett rumsligt temporalt neuralt nätverk (ST-GRU) föreslagits. Den består av två delar, nämligen ST och GRU. Den första delen är en extraktionsalgoritm som drar ut relevanta korrelationer mellan tid och plats som är inkorporerade i nätverket. Den andra delen, GRU, förutspår nästa plats med avseende på användarens aktuella plats. Studien visar att den föreslagna modellen ST-GRU ger bättre resultat jämfört med benchmarkmodellerna.

Made available in DSpace on 2015-04-22T22:00:46Z (GMT). No. of bitstreams: 1 Pamela Campos Levy.pdf: 4863540 bytes, checksum: 820e34768b005399acf73dec3e491ae5 (MD5) Previous issue date: 2012-08-24 / FAPEAM - Fundação de Amparo à Pesquisa do Estado do Amazonas / Tuberculosis (TB) is an infectious disease transmitted by Koch's bacillus, or Mycobacterium tuberculosis. An estimated 1.4 million people died of tuberculosis in 2010. About 95% of these deaths occurred in developing countries, or development. In Brazil, each year are registered more than 68,000 new cases. Currently, Amazon is the Brazilian state with the highest incidence rate of the disease. a of TB diagnostic methods, adopted by the Ministry of Health is examining smear of bright field. The smear is the count of bacilli in slides containing sputum samples of the patient, prepared and stained according to the methodology standard. Over the past five years, research related to the recognition of bacilli tuberculosis, using images obtained by microscopy bright field, has been carried out with a view to automating this diagnostic method, given the fact that the number high smear tests performed by professional induce eyestrain and due to diagnostic errors. This paper presents a new method of recognition and targeting of tubercle bacilli in slides fields of images, containing pulmonary secretions of the patient, stained by Kinyoun method. From these bacilli images of pixels and background samples were extracted for training classifier. Images were automatically broken down into two groups, according with substantial content. The developed method selects an optimal set of color characteristics of the bacillus and of the background, using the method of selection climbing characteristics. These features were used in a pixel classifier, a multilayer perceptron, trained by backpropagation algorithm. The optimal set of features selected, {GI, Y-Cr, La, RG, a}, from the RGB color spaces, HSI, YCbCr and Lab, combined with the network perceptron with eighteen (18) neurons in first layer three (3) and the second one (1) in the third (18-3-1), resulted in an accuracy of 92.47% in the segmentation of bacilli. The image discrimination method in relation to automated background content contributed to affirm that the method described in this paper it is more appropriate to target bacilli images with low content density background (more uniform background). For future work, new techniques to remove noise present in images with high density of background content (containing background many artifacts) should be developed. / A tuberculose (TB) é uma doença infectocontagiosa, transmitida pelo bacilo de Koch, ou Mycobacterium tuberculosis. Estima-se que 1,4 milhões de pessoas morreram de tuberculose em 2010. Cerca de 95% dessas mortes ocorreram em países subdesenvolvidos ou em desenvolvimento. No Brasil, a cada ano são registrados mais de 68 mil novos casos. Atualmente, o Amazonas é o estado brasileiro com a maior taxa de incidência da doença. Um dos métodos de diagnóstico da TB, adotado pelo Ministério da Saúde, é o exame de baciloscopia de campo claro. A baciloscopia consiste na contagem dos bacilos em lâminas contendo amostras de escarro do paciente, preparadas e coradas de acordo com metodologia padronizada. Nos últimos cinco anos, pesquisas relacionadas ao reconhecimento de bacilos da tuberculose, utilizando imagens obtidas por microscopia de campo claro, tem sido realizadas com vistas a automatização desse método diagnóstico, em face do fato de que o número elevado de exames de baciloscopia realizado pelos profissionais induzirem a fadiga visual e em consequência a erros diagnósticos. Esse trabalho apresenta um novo método de reconhecimento e segmentação de bacilos da tuberculose em imagens de campos de lâminas, contendo secreção pulmonar do paciente, coradas pelo método de Kinyoun. A partir dessas imagens foram extraídas amostras de pixels de bacilos e de fundo para treinamento do classificador. As imagens foram automaticamente discriminadas em dois grupos, de acordo com o conteúdo de fundo. O método desenvolvido seleciona um conjunto ótimo de características de cor do bacilo e do fundo da imagem, empregando o método de seleção escalar de características. Essas características foram utilizadas em um classificador de pixels, um perceptron multicamada, treinado pelo algoritmo backpropagation. O conjunto ótimo de características selecionadas, {G-I, Y-Cr, L-a, R-G, a}, proveniente dos espaços de cores RGB, HSI, YCbCr e Lab, combinado com a rede perceptron com 18 (dezoito) neurônios na primeira camada, 3 (três) na segunda e 1 (um) na terceira (18-3-1), resultou em uma acurácia de 92,47% na segmentação dos bacilos. O método de discriminação de imagens em relação ao conteúdo de fundo automatizado contribuiu para afirmar que o método descrito neste trabalho é mais adequado para segmentar bacilos em imagens com baixa densidade de conteúdo de fundo (fundo mais uniforme). Para os trabalhos futuros, novas técnicas para remover os ruídos presentes em imagens com alta densidade de conteúdo de fundo (fundo contendo muitos artefatos) devem ser desenvolvidas.