In recent years, deep neural networks have attracted lots of attentions in the field of computer vision and artificial intelligence. Convolutional neural network exploits spatial correlations in an input image by performing convolution operations in local receptive fields. When compared with fully connected neural networks, convolutional neural networks have fewer weights and are faster to train. Many research works have been conducted to further reduce computational complexity and memory requirements of convolutional neural networks, to make it applicable to low-power embedded applications. This thesis focuses on a special class of convolutional neural network with only binary weights and activations, referred as binary neural networks. Weights and activations for convolutional and fully connected layers are binarized to take only two values, +1 and -1. Therefore, the computations and memory requirement have been reduced significantly. The proposed architecture of binary neural networks has been implemented on an FPGA as a real time, high speed, low power computer vision platform. Only on-chip memories are utilized in the FPGA design. The FPGA implementation is evaluated using the CIFAR-10 benchmark and achieved a processing speed of 332,164 images per second for CIFAR-10 dataset with classification accuracy of about 86.06%.

Too many traffic accidents are caused by drivers’ failure of noticing buildings, traffic sign and other objects. Video based scene or object detection which can easily enhance drivers’ judgment performance by automatically detecting scene and signs. Two of the recent popular video detection algorithms are Background Differentiation and Feature based object detection. The background Differentiation is an efficient and fast way of observing a moving object in a relatively stationary background, which makes it easy to be implemented on a mobile platform and performs a swift processing speed. The Feature based scene detection such like the Speeded Up Robust Feature (SURF), is an appropriate way of detecting specific scene with accuracy and rotation and illumination invariance. By comparison, SURF computational expense is much higher, which remains the algorithm limited in real time mobile platform. In this thesis, I present two real time tracking algorithms, Differentiation based and SURF based scene detection systems on FPGA platform. The proposed hardware designs are able to process video of 800*600 resolution at 60 frames per second, the video clock rate is 40 MHz.

Software used in medical settings operate in complex and variable environments. Programs need to integrate well not only with their electrical and mechanical components, but also within the socio-technological setting they participate in. In this Master's Thesis, a modular software architecture for controlling surgical robot systems within magnetic resonance scanners is designed and implemented. The C++ program runs on a sbRIO 9651 real-time operating system and an object oriented design is taken. Robot kinematics and controls are put into effect in software and validated. Communication with up to ten daughter cards occurs via SPI and external information is exchanged via OpenIGTLink. A web-based engineering console made with ReactJS is also constructed to provide a visual interface for actuating motor axes and executing robot functionality. Documentation of the code is provided and the program was validated quantitatively with software tests and qualitatively through experimentation in MRI suites.

Early time series classification is the task of predicting the class label of a time series before it is observed in its entirety. In time-sensitive domains where information is collected over time it is worth sacrificing some classification accuracy in favor of earlier predictions, ideally early enough for actions to be taken. However, since accuracy and earliness are contradictory objectives, a solution to this problem must find a task-dependent trade-off. There are two common state-of-the-art methods. The first involves an analyst selecting a timestep at which all predictions must be made. This does not capture earliness on a case-by-case basis, so if the selecting timestep is too early, all later signals are missed, and if a signal happens early, the classifier still waits to generate a prediction. The second method is the exhaustive search for signals, which encodes no timing information and is not scalable to high dimensions or long time series. We design the first early classification model called EARLIEST to tackle this multi-objective optimization problem, jointly learning (1) to decide at which time step to halt and generate predictions and (2) how to classify the time series. Each of these is learned based on the task and data features. We achieve an analyst-controlled balance between the goals of earliness and accuracy by pairing a recurrent neural network that learns to classify time series as a supervised learning task with a stochastic controller network that learns a halting-policy as a reinforcement learning task. The halting-policy dictates sequential decisions, one per timestep, of whether or not to halt the recurrent neural network and classify the time series early. This pairing of networks optimizes a global objective function that incorporates both earliness and accuracy. We validate our method via critical clinical prediction tasks in the MIMIC III database from the Beth Israel Deaconess Medical Center along with another publicly available time series classification dataset. We show that EARLIEST out-performs two state-of-the-art LSTM-based early classification methods. Additionally, we dig deeper into our model's performance using a synthetic dataset which shows that EARLIEST learns to halt when it observes signals without having explicit access to signal locations. The contributions of this work are three-fold. First, our method is the first neural network-based solution to early classification of time series, bringing the recent successes of deep learning to this problem. Second, we present the first reinforcement-learning based solution to the unsupervised nature of early classification, learning the underlying distributions of signals without access to this information through trial and error. Third, we propose the first joint-optimization of earliness and accuracy, allowing learning of complex relationships between these contradictory goals.

PURPOSE: The aim of this study is to better characterize the effect of the interval in time to the operating room on hospital length of stay and other post-surgical outcomes in adult patients with common emergency general surgery conditions who are admitted to the acute care surgical service at Boston Medical Center. METHODS: This is retrospective cohort study examining a total of 321 subjects taken from an emergency general surgery registry at Boston Medical Center from May 2014 thru May 2015. Variables analyzed included: demographic factors, Charlson Comorbidity Index scores, times to the operating room, hospital length of stays, post- operative complications, and in-hospital mortality. RESULTS: There were zero mortalities in this study and a 3.1% post-operative complication rate. There was a positive association between time to the operating room and hospital length of stay, even after controlling for covariates. It was found that those subjects who go to the operating room after 6 hours from the time of admission have an increased hospital length of stay by about 12 hours as compared to those subjects who do not. CONCLUSIONS: In this study, subjects who went to the operating room sooner from the time of admission had associated shorter hospital length of stays and fewer post- operative complications.

This dissertation presents new numerical methods for the solution of fractional differential equations of single and distributed order that find application in the different fields of physics and engineering. We start by presenting the relationship between fractional derivatives and processes like anomalous diffusion, and, we then develop new numerical methods for the solution of the time-fractional diffusion equations. The first numerical method is developed for the solution of the fractional diffusion equations with Neumann boundary conditions and the diffusivity parameter depending on the space variable. The method is based on finite differences, and, we prove its convergence (convergence order of O(Δx² + Δt²^-α), 0 < α < 1) and stability. We also present a brief description of the application of such boundary conditions and fractional model to real world problems (heat flux in human skin). A discussion on the common substitution of the classical derivative by a fractional derivative is also performed, using as an example the temperature equation. Numerical methods for the solution of fractional differential equations are more difficult to develop when compared to the classical integer-order case, and, this is due to potential singularities of the solution and to the nonlocal properties of the fractional differential operators that lead to numerical methods that are computationally demanding. We then study a more complex type of equations: distributed order fractional differential equations where we intend to overcome the second problem on the numerical approximation of fractional differential equations mentioned above. These equations allow the modelling of more complex anomalous diffusion processes, and can be viewed as a continuous sum of weighted fractional derivatives. Since the numerical solution of distributed order fractional differential equations based on finite differences is very time consuming, we develop a new numerical method for the solution of the distributed order fractional differential equations based on Chebyshev polynomials and present for the first time a detailed study on the convergence of the method. The third numerical method proposed in this thesis aims to overcome both problems on the numerical approximation of fractional differential equations. We start by solving the problem of potential singularities in the solution by presenting a method based on a non-polynomial approximation of the solution. We use the method of lines for the numerical approximation of the fractional diffusion equation, by proceeding in two separate steps: first, spatial derivatives are approximated using finite differences; second, the resulting system of semi-discrete ordinary differential equations in the initial value variable is integrated in time with a non-polynomial collocation method. This numerical method is further improved by considering graded meshes and an hybrid approximation of the solution by considering a non-polynomial approximation in the first sub-interval which contains the origin in time (the point where the solution may be singular) and a polynomial approximation in the remaining intervals. This way we obtain a method that allows a faster numerical solution of fractional differential equations (than the method obtained with non-polynomial approximation) and also takes into account the potential singularity of the solution. The thesis ends with the main conclusions and a discussion on the main topics presented along the text, together with a proposal of future work.

Esta dissertação apresenta aspectos relacionados às instituições de ensino particulares na cidade de Porto Alegre, capital do Rio Grande do Sul, entre 1927 e 1957. Tempo, espaço e cidade são fatores que fazem parte da análise que envolve escolarização e urbanização da capital sul-rio-grandense. Visando identificar características das escolas fundadas na cidade, optou-se por fazer um levantamento dos educandários e suas particularidades, além da localização e possíveis alteração de endereço. Por meio da pesquisa em livros, trabalhos acadêmicos, artigos, textos e materiais históricos, como por exemplo, documentações oficiais, relatórios de instrução e de intendência, atas, documentos municipais e estaduais, legislações, mapas e através da internet, foi efetuado o levantamento das escolas e sua localização nos mapas de 1928 e 1952. Visitas aos acervos de bibliotecas, arquivos, consultas virtuais foram realizadas para obtenção dos dados e posterior análise das informações e na relação destas com características socioeconômicas, políticas, culturais, étnicas e confessionais. Após examinar o conjunto de informações foi possível perceber aspectos que impactaram nos processos de escolarização do período da pesquisa. Cynthia Greive Veiga, Luciano Faria Filho, Milton Santos e Célia Ferraz de Souza foram autores que colaboraram com as análises e reflexões deste estudo. / This thesis presents aspects related to the private education institutions in Porto Alegre, the capital of Rio Grande do Sul, between 1927 and 1957. Time, space and city are factors involved in the analysis that concerns schooling and urbanization in the capital of Rio Grande do Sul. Aiming to identify the characteristics of the schools established in the city, it was chosen to make a survey of the schools and their features, as well as their location and any possible change of address. Through research in books, academic works, articles, texts and historical materials, such as official documentation, instruction and stewardship reports, minutes, municipal and state documents, legislations, maps and the internet use, a survey was carried out on the schools and their locations in the 1928 and 1952 maps. Archives and libraries collections were visited and virtual consultations were carried out to obtain the data, later information analysis and their relation with socioeconomic, political, cultural, ethnic and confessional characteristics. After examining the set of information, it was possible to detect aspects that impacted the schooling processes of the research period. Cynthia Greive Veiga, Luciano Faria Filho, Milton Santos and Célia Ferraz de Souza were authors who collaborated with the analyzes and reflections of this study.

Typescript (photocopy). / Digitized by Kansas Correctional Industries

Time is a critical element for the understanding of natural processes (e.g., earthquakes and weather) or human-made artifacts (e.g., stock market and speech signals). The analysis of time series, the result of sequentially collecting observations of such processes and artifacts, is becoming increasingly prevalent across scientific and industrial applications. The extraction of non-trivial features (e.g., patterns, correlations, and trends) in time series is a critical step for devising effective time-series mining methods for real-world problems and the subject of active research for decades. In this dissertation, we address this fundamental problem by studying and presenting computational methods for efficient unsupervised learning of robust feature representations from time series. Our objective is to (i) simplify and unify the design of scalable and accurate time-series mining algorithms; and (ii) provide a set of readily available tools for effective time-series analysis. We focus on applications operating solely over time-series collections and on applications where the analysis of time series complements the analysis of other types of data, such as text and graphs. For applications operating solely over time-series collections, we propose a generic computational framework, GRAIL, to learn low-dimensional representations that natively preserve the invariances offered by a given time-series comparison method. GRAIL represents a departure from classic approaches in the time-series literature where representation methods are agnostic to the similarity function used in subsequent learning processes. GRAIL relies on the attractive idea that once we construct the data-to-data similarity matrix most time-series mining tasks can be trivially solved. To overcome scalability issues associated with approaches relying on such matrices, GRAIL exploits time-series clustering to construct a small set of landmark time series and learns representations to reduce the data-to-data matrix to a data-to-landmark points matrix. To demonstrate the effectiveness of GRAIL, we first present domain-independent, highly accurate, and scalable time-series clustering methods to facilitate exploration and summarization of time-series collections. Then, we show that GRAIL representations, when combined with suitable methods, significantly outperform, in terms of efficiency and accuracy, state-of-the-art methods in major time-series mining tasks, such as querying, clustering, classification, sampling, and visualization. Overall, GRAIL rises as a new primitive for highly accurate, yet scalable, time-series analysis. For applications where the analysis of time series complements the analysis of other types of data, such as text and graphs, we propose generic, simple, and lightweight methodologies to learn features from time-varying measurements. Such applications often organize operations over different types of data in a pipeline such that one operation provides input---in the form of feature vectors---to subsequent operations. To reason about the temporal patterns and trends in the underlying features, we need to (i) track the evolution of features over different time periods; and (ii) transform these time-varying features into actionable knowledge (e.g., forecasting an outcome). To address this challenging problem, we propose principled approaches to model time-varying features and study two large-scale, real-world, applications. Specifically, we first study the problem of predicting the impact of scientific concepts through temporal analysis of characteristics extracted from the metadata and full text of scientific articles. Then, we explore the promise of harnessing temporal patterns in behavioral signals extracted from web search engine logs for early detection of devastating diseases. In both applications, combinations of features with time-series relevant features yielded the greatest impact than any other indicator considered in our analysis. We believe that our simple methodology, along with the interesting domain-specific findings that our work revealed, will motivate new studies across different scientific and industrial settings.

Trata-se de um estudo de natureza exploratória descritiva, com abordagem qualitativa. O arcabouço teórico que subsidiou a construção dos pressupostos e a análise dos resultados foram problematizações da Nova História francesa, aliadas às contribuições da fenomenologia sobre o tempo vivido, visando à exploração histórica e à compreensão do fenômeno estudado. Quanto às técnicas ou procedimentos de pesquisa, utilizamos a história oral temática acerca do tempo no hospital psiquiátrico através da questão norteadora: “Como você registra o tempo aqui no Hospital?" e suas correlatas, “Há quanto tempo você está no Hospital?", “Como você passa o dia no Hospital?". Objetivamos investigar as estruturas de sustentação temporal de pacientes psiquiátricos indigentes, com história de mais de vinte anos de internação em manicômio, que não dispunham de calendários e/ou relógios, assim como, relacionarmos tais estruturas às práticas e aos agentes da cultura manicomial e relacionarmos a negação do tempo no hospital psiquiátrico tradicional à possibilidade de acesso ventilada pela transformação dos serviços de atenção em saúde mental (a instituição negada/inventada). Os resultados encontrados apontam para a predominância de um tempo que, embora “morto", se impõe na ausência de conectores convencionais (calendário e relógio), através de nostalgia, silêncio eloqüente, um distanciamento que oscila entre uma imediata justificação ou “aversão" ao tema. Os sujeitos da pesquisa denunciam uma temporalidade interceptada, uma “presença faltante" marcada pela vivência institucional, mas também por resquícios de um passado pessoal, onde o porvir, hoje, propugnado pela atenção psicossocial exige o “re-aprendizado" das formas de lidar com os mecanismos temporais que caracterizam a passagem do tempo na sociedade extramuro. / This is an exploratory-descriptive study using a qualitative approach. The framework that based the construction of hypothesis and results analysis were the contributions of French New History and phenomenology, enabling a history exploration and the understanding of the studied phenomenon. With respect to the research techniques and procedures, the author used oral history to analyze time at a psychiatric hospital through the guiding question: “ How do you feel time at the hospital?" and others related questions such as “How long have you been here?" , “How is your day at the hospital?". The aim of this study was to investigate the structures that base time to psychiatric patients with a history of more than twenty years of hospitalization, who did not have calendars and/or watches, as well as to relate these structures to practices, to the traditional hospital culture of denying time and to the possibility of access and changes in mental health services (from denied to an invented institution). The results point out the predominance of time, that although “dead", is present even with the lack of conventional connectors (calendars and watches), through nostalgia, eloquent silence, a distance that oscillates from an immediate justification to an “aversion" regarding the theme. The subjects of this research denounced an interrupted temporality, a “lack" marked by their institutional experiences but also with elements of their past, in which tomorrow and today, that are the basis of psychosocial care require the re-learning of coping ways with temporal mechanisms that characterize time flow in outdoors society.