[pt] O sistema operacional MINIX é uma nova implementação do sistema UNIX, versão 7, feito para fins didáticos. Ele está formado por uma coleção de processos, estruturados em 4 niveis: administração de processos, processos básicos do sistema, processos servidores de memória e arquivos, e processos usuários. A presente dissertação descreve o transporte do sistema MINIX, do microcomputador IBM PC XT para o computador CYGNUS do laboratório de Sistema de Computação do Departamento de Engenharia Elétrica da PUC/RJ, o qual possui uma arquitetura baseada em processadores da linha Motorola (68010 e 68020). O trabalho do transporte, constituiu em adaptar o hardware do CYGNUS para receber o sistema operacional MINIX, reprojetar o MINIX para o hardware do CYGNUS, e transportar o MINIX em forma completa, para seu novo ambiente de execução. / [en] The MINIX operating System is a new implementation of the UNIX operating system version 7, designed for didactic purposes. It is arranged as a collection of processes, structured in 4 levels: process management, system basic processes, memory and file server processes, and user processes. This work describes the transport of the MINIX system from the IBM PC XT microcomputer to te CYGNUS computer, which was developed in the Computer System Laboratory of the Electrical Engineering Department of PUC/RJ, with an architecture based on Motorola processors (68010 and 68020). The task of transporting the system, consisted in adjusting the CYCNUS hardware to accept the MINIX, redesigning the MINIX for the CYGNUS hardware, and finaly, in transporting the MINIX in its complete form to its new execution environment.

Exercise is a prevailing topic in modern society as more people are pursuing a healthy lifestyle. Physical activities provide unimaginable benefits to human well-being from the inside out. 2D human pose estimation, action recognition and repetitive counting fields developed rapidly in the past several years. However, few works combined them together as a whole system to assist people in evaluating body poses, recognizing exercises and counting repetitive actions. The existing methods estimate pose positions first, and utilize human joints locations in the other two tasks. In this thesis, we propose a multitask system covering the three domains. Different from the methodology used in the literature, heatmaps which are the byproducts of 2D human pose estimation models are adopted for exercise recognition and counting. Recent heatmap processing methods are proven effective in extracting dynamic body pose information. Inspired by this, we propose a new deep-learning multitask model of exercise recognition & repetition counting, and apply these approaches to the multitask for the first time. To meet the needs of the multitask model, we create a new dataset Rep-Penn with action, counting and speed labels. A two-stage training strategy is applied in the training process. Our multitask system can estimate human pose, identify physical activities and count repeated motions. We achieved 95.69% accuracy in exercise recognition on Rep-Penn dataset. The multitask model also performed well in repetitive counting with 0.004 Mean Average Error (MAE) and 0.997 Off-By-One (OBO) accuracy on Rep-Penn dataset. Compared with existing frameworks, our method obtained state-of-the-art results.

<p>In community mobility, walking is commonly completed with other concurrent tasks, described as locomotor multitasks. Many locomotor multitasks rely on vision for both gait and concurrent tasks. When each of the individual tasks uses the same perceptual modality (e.g. vision), structural interference occurs. Structural interference is different from capacity interference, which refers to tasks competing for limited cognitive resources. While locomotor multitask studies have demonstrated that completing the locomotor multitask typically leads to performance impairment in gait and/or the concurrent task, the wide range of tasks has confounded the ability to fully understand how structural and capacity interference affect multitask performance. Thus, the purpose of this dissertation was to delineate how structural interference (Study 1) and capacity interference (Study 2) affect gait multitask performance. To facilitate comparison across studies, the two studies (Study 1 and Study 2) in this dissertation used the same gait task – obstacle crossing – and the same cognitive task – a visual discrete reaction time (RT) task. A discrete RT task was completed while approaching to an obstacle, where visual information regarding obstacle is being gathered to plan for the successful obstacle crossing. In Study 1, to determine if structural interference affects performance impairment in young and older adults, gaze diversion was manipulated by the RT task location (gaze diverted to the obstacle, and gaze diverted away from the obstacle). The RT task was also completed while standing to strengthen the interpretation that any performance impairments were due to structural interference. Study 1 results indicated that structural interference affects both gait and cognitive task performance. Structural interference demonstrated performance impairments in both young and older adults, but the strategies were different. Young adults were more likely adopt gait behavior that increased the risk of tripping when gaze was diverted away from the obstacle (high structural interference), but older adults demonstrated a strategy that decreased the risk of trip when gaze was diverted to the obstacle (low structural interference). This finding highlights the critical role of vision in adaptive gait. In study 2, to determine if capacity interference affects performance impairment in young adults, both gait and cognitive task were manipulated while structural interference was held constant; gait task was manipulated by obstacle height (level walking, 15% leg length height, and 30% leg length height obstacle), and cognitive tasks were three RT tasks (Simple RT, Choice RT, Simon RT). The baseline for each gait task (without RT task) and cognitive task (while seating) was also measured. Capacity interference demonstrated that task prioritization strategy was different for gait challenge versus cognitive challenge in young adults. As gait task difficulty increased, gait task was prioritized. Conversely, as cognitive task difficulty increased, cognitive task was prioritized. This finding highlights that young adults have the ability to flexibly allocate the resources to accomplish the multitask. Lastly, an interesting finding from two studies (Study 1 and Study 2) was when interference is applied during the planning phase – during the approach to the obstacle – structural interference has a greater effect on obstacle crossing performance than capacity interference.</p>

We explore how to automatically detect specific phrases in audio from noisy, multi-speaker videos using deep neural networks. Specifically, we focus on classroom observation videos that contain a few adult teachers and several small children (< 5 years old). At any point in these videos, multiple people may be talking, shouting, crying, or singing simultaneously. Our goal is to recognize polite speech phrases such as "Good job", "Thank you", "Please", and "You're welcome", as the occurrence of such speech is one of the behavioral markers used in classroom observation coding via the Classroom Assessment Scoring System (CLASS) protocol. Commercial speech recognition services such as Google Cloud Speech are impractical because of data privacy concerns. Therefore, we train and test our own custom models using a combination of publicly available classroom videos from YouTube, as well as a private dataset of real classroom observation videos collected by our colleagues at the University of Virginia. We also crowdsource an additional 1152 recordings of polite speech phrases to augment our training dataset. Our contributions are the following: (1) we design a crowdsourcing task for efficiently labeling speech events in classroom videos, (2) we develop a neural network-based architecture for speech recognition, robust to noise and overlapping speech, and (3) we explore methods to synthesize new and authentic audio data, both to increase the training set size and reduce the class imbalance. Finally, using our trained polite speech detector, (4) we investigate the relationship between polite speech and CLASS scores and enable teachers to visualize their use of polite language.

This dissertation addresses probabilistic inference via relative entropy minimization subject to expectation constraints. A canonical representation of the solution is determined without the requirement for convexity of the constraint set, and is given by members of an exponential family. The use of conjugate priors for relative entropy minimization is proposed, and a class of conjugate prior distributions is introduced. An alternative representation of the solution is provided as members of the prior family when the prior distribution is conjugate. It is shown that the solutions can be found by direct optimization with respect to members of such parametric families. Constrained Bayesian inference is recovered as a special case with a specific choice of constraints induced by observed data. The framework is applied to the development of novel probabilistic models for multitask learning subject to constraints determined by domain expertise. First, a model is developed for multitask learning that jointly learns a low rank weight matrix and the prior covariance structure between different tasks. The multitask learning approach is extended to a class of nonparametric statistical models for transposable data, incorporating side information such as graphs that describe inter-row and inter-column similarity. The resulting model combines a matrix-variate Gaussian process prior with inference subject to nuclear norm expectation constraints. In addition, a novel nonparametric model is proposed for multitask bipartite ranking. The proposed model combines a hierarchical matrix-variate Gaussian process prior with inference subject to ordering constraints and nuclear norm constraints, and is applied to disease gene prioritization. In many of these applications, the solution is found to be unique. Experimental results show substantial performance improvements as compared to strong baseline models. / text

No description available.

Algorithms that learn sets of rules describing a concept from its examples have been widely studied in machine learning and have been applied to problems in medicine, molecular biology, planning and linguistics. Many of these algorithms used a separate-and-conquer strategy, repeatedly searching for rules that explain different parts of the example set. When examples are scarce, however, it is difficult for these algorithms to evaluate the relative quality of two or more rules which fit the examples equally well. This dissertation proposes, implements and examines a general technique for modifying rule evaluation in order to improve learning performance in these situations. This approach, called Description-based Evaluation Function Transfer (DEFT), adjusts the way rules are evaluated on a target concept by taking into account the performance of similar rules on a related support task that is supplied by a domain expert. Central to this approach is a novel theory of task similarity that is defined in terms of syntactic properties of rules, called descriptions, which define what it means for rules to be similar. Each description is associated with a prior distribution over classification probabilities derived from the support examples and a rule's evaluation on a target task is combined with the relevant prior using Bayes' rule. Given some natural conditions regarding the similarity of the target and support task, it is shown that modifying rule evaluation in this way is guaranteed to improve estimates of the true classification probabilities. Algorithms to efficiently implement Deft are described, analysed and used to measure the effect these improvements have on the quality of induced theories. Empirical studies of this implementation were carried out on two artificial and two real-world domains. The results show that the inductive transfer of evaluation bias based on rule similarity is an effective and practical way to improve learning when training examples are limited.

Algorithms that learn sets of rules describing a concept from its examples have been widely studied in machine learning and have been applied to problems in medicine, molecular biology, planning and linguistics. Many of these algorithms used a separate-and-conquer strategy, repeatedly searching for rules that explain different parts of the example set. When examples are scarce, however, it is difficult for these algorithms to evaluate the relative quality of two or more rules which fit the examples equally well. This dissertation proposes, implements and examines a general technique for modifying rule evaluation in order to improve learning performance in these situations. This approach, called Description-based Evaluation Function Transfer (DEFT), adjusts the way rules are evaluated on a target concept by taking into account the performance of similar rules on a related support task that is supplied by a domain expert. Central to this approach is a novel theory of task similarity that is defined in terms of syntactic properties of rules, called descriptions, which define what it means for rules to be similar. Each description is associated with a prior distribution over classification probabilities derived from the support examples and a rule's evaluation on a target task is combined with the relevant prior using Bayes' rule. Given some natural conditions regarding the similarity of the target and support task, it is shown that modifying rule evaluation in this way is guaranteed to improve estimates of the true classification probabilities. Algorithms to efficiently implement Deft are described, analysed and used to measure the effect these improvements have on the quality of induced theories. Empirical studies of this implementation were carried out on two artificial and two real-world domains. The results show that the inductive transfer of evaluation bias based on rule similarity is an effective and practical way to improve learning when training examples are limited.

Cada vez mais equipamentos eletrônicos digitais têm sido fabricados utilizando um sistema operacional embarcado. Por razões de custo, estes sistemas operacionais são implementados sobre um hardware com os requisitos mínimos para atender as necessidades da aplicação. Este trabalho apresenta um estudo sobre a viabilidade de implementação de suporte a multitarefa sobre a arquitetura FemtoJava, um microcontrolador monotarefa dedicado a sistemas embarcados. Para tanto, o suporte de hardware necessário é adicionado à arquitetura. Também são implementados dois escalonadores de tarefas diretamente em bytecodes Java, visando à otimização de área e o compromisso com desempenho e consumo de energia. Modificações no ambiente de desenvolvimento e uma ferramenta de relocação de endereços são propostas, objetivando a utilização dos escalonadores de tarefas implementados junto ao fluxo de desenvolvimento existente. Por fim, uma análise é realizada sobre o impacto que a capacidade de multitarefa produz no sistema em termos de desempenho, consumo de área e energia. / Most digital electronic equipments are produced using an embedded operating system. Due to economic reasons, these operating systems are implemented on hardware with minimal requirements to support the application needs. This work will present a viability study to implement multitask support on the FemtoJava architecture, a monotask microcontroller dedicated to embedded applications. The support to multitask involves the addition of specific hardware mechanisms to the architecture. Two different scheduling policies are then directly implemented using Java bytecodes, aiming area optimization as well as a good performance/energy-consumption trade-off. Some modifications in the development environment and a code relocation tool were introduced, in order to enable the use of the schedulers in the existing design tool flow. Finally, an analysis is performed to evaluate the impact that the multitask support produces in the system with respect to the final performance, area and energy consumption.

Submitted by Roberto Zeitounlian Sarkisian (roberto.sarkisian@fgv.br) on 2012-03-14T21:19:40Z No. of bitstreams: 1 Roberto Zeitounlian Sarkisian - Dissertação Mestrado EESP.pdf: 342262 bytes, checksum: b62310824eb0781e1a5a3cf541fde0ac (MD5) / Approved for entry into archive by Gisele Isaura Hannickel (gisele.hannickel@fgv.br) on 2012-03-15T14:23:07Z (GMT) No. of bitstreams: 1 Roberto Zeitounlian Sarkisian - Dissertação Mestrado EESP.pdf: 342262 bytes, checksum: b62310824eb0781e1a5a3cf541fde0ac (MD5) / Made available in DSpace on 2012-03-15T15:13:25Z (GMT). No. of bitstreams: 1 Roberto Zeitounlian Sarkisian - Dissertação Mestrado EESP.pdf: 342262 bytes, checksum: b62310824eb0781e1a5a3cf541fde0ac (MD5) Previous issue date: 2012-02-13 / A central hipothesis in a principal-agent relation is that the principal always seeks to maximize profits. This paper studies incentives provision when the principal has multiple objectives besides profit maximization and the additional performance measure is costly to observe. The model used is a modification of the multitask model presented by Holmstr¨om and Milgrom (1991). As results, we find that the optimal contract changes accordingly to the correlation coefficient between the noise terms in the performance measurements. Besides that, not only a principal whose main concern is the second task, as a principal who has clear preferences for product, will willingly pay the observability cost of the additional performance measure when the correlation coefficient is positive. / Uma hipótese central nos modelos de relação agente-principal é a de que o principal busca maximizar o lucro. O objetivo do trabalho é estudar a provisão de incentivos quando o principal possui múltiplos objetivos além da maximização do lucro e o uso de medidas adicionais de performance é custoso. O modelo apresentado é uma modificação do modelo de multitask proposto por Holmström e Milgrom (1991). Como resultados, observa-se que a forma do contrato ótimo é sensível à correlação entre os ruídos das medidas de performance e que tanto um principal que se preocupa muito com o objetivo adicional, quanto um principal que se preocupa muito com o produto da firma, estarão dispostos a pagar o custo de observação da medida de performance adicional quando a correlação entre os ruídos é positiva.