Thesis (Ph.D)--Computing, Georgia Institute of Technology, 2010. / Committee Chair: Rehg, James M.; Committee Member: Christensen, Henrik; Committee Member: Dellaert, Frank; Committee Member: Essa, Irfan; Committee Member: Malik, Jitendra. Part of the SMARTech Electronic Thesis and Dissertation Collection.

Mutation testing and adequacy testing are two major technologies to assure the quality of software. In this thesis, we present the first work that alleviates the high cost of mutation testing and ineffectiveness of adequacy testing for context-aware applications. We also present large-scale multi-subject case studies to evaluate how our work successfully alleviates these problems. Mutation testing incurs a high execution cost if randomly selected test inputs kill a small percentage of remaining live mutants. To address this problem, we formulate the notion of context diversity to measure the context changes inherent in test inputs, and propose three context-aware strategies in the selection of test inputs. The empirical results show that the use of test inputs with higher context diversity can significantly benefit mutation testing in terms of resulting in fewer test runs, fewer test case trials, and smaller resultant test suites that achieve a high mutation score level. The case study also shows that at the test case level, the context diversity of test inputs positively and strongly correlates with multiple types of adequacy metrics, which provide a foundation on why context diversity contributes to the effectiveness of test cases in revealing faults in context-aware applications. In adequacy testing, many strategies randomly select test cases to construct adequate test suites with respect to program-based adequacy criteria. They usually exclude redundant test cases that are unable to improve the coverage of the test requirements of an adequacy criterion achieved by constructing test suites. These strategies have not explored in the diversity in test inputs to improve the test effectiveness of test suites. To address this problem, we propose three context-aware refined strategies to check whether redundant test cases can replace previously selected test cases to achieve the same coverage level but with different context diversity levels. The empirical study shows that context diversity can be significantly injected into adequate test suites, and favoring test cases with higher context diversity can significantly improve the fault detection rates of adequate test suites for testing context-aware applications. In conclusion, this thesis makes the significant contributions to the research in testing context-aware applications: (1) It has formulated context diversity, a novel metric to measure context changes inherent in test inputs. (2) It has proposed three context-aware strategies to select test cases with different levels of context diversity. Compared with the baseline strategy, the strategy CAS-H that uses test cases with higher context diversity can significantly reduce the cost of mutation testing over context-aware applications in terms of less number of test runs, smaller adequate test suites, and less number of test inputs used to construct test suites. (3) It has defined three context-aware refined strategies to construct adequate test suites with different context diversity levels. Compared with the baseline strategy, the strategy CARS-H that favors test cases with higher context diversity can significantly improve the effectiveness of adequacy testing in terms of higher fault detection rates. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

The allocation of resources to tasks in a computationally efficient manner is a key problem in computer science. One important application domain for solutions to this class of problem is the allocation of sensing resources for environmental monitoring, surveillance, or similar sensing tasks. Within this domain, however, the complexity of the problem is compounded by a number of factors: new tasks may arrive at any time, resources may be shared between tasks under some conditions, tasks may be composed of inter-dependent sub-tasks, and tasks may compete for sensor resources. These factors combined with the dynamic nature of the topology of sensor networks (e.g. sensors may move out of range or become damaged) mean that it is extremely difficult or impossible to have a solution using existing techniques. In this thesis, we propose an efficient, agent-based solution (FAST for Flexible Allocation for Sensing Tasks) to this complex dynamic problem. The sensing resources in FAST can be either static or mobile or a mixture of both. Particularly, each resource is managed by a task leader agent (i.e. the actual sensor that is closest to the task central point). The problem is then modelled as a coordination problem where the task agents employ a novel multi-round Knapsack-based algorithm (GAP-E) to obtain a solution. If there are dependencies between sub-tasks, such relationships are solved prior to the actual allocation. At execution time, if there is any environment change that affects the task sensing type requirements, the previously determined sensor types for tasks are revised. When applicable, the agents are cooperative through exchanging and sharing resources to maximise their profits. In addition, FAST addresses the situation where sensor resource sharing is not possible and there is no incentive for sensor resources to be exchanged. In such situations, an additional post-process step underpinned by mechanism for exchanging resources through negotiation were introduced. Through those mechanisms, agents may, in a decentralized manner, decide the means to deliver on a sensing task given local conditions, and to alleviate the impact of task arrival time on the quality of the global solution. Via empirical evaluation, these steps significantly improved the number of sensing tasks that can be successfully completed with only a minor impact on execution time.

Au cours des dernières années, le nombre de systèmes informatiques n'a pas cesser d'augmenter. Les centres de données sont peu à peu devenus des équipements hautement demandés et font partie des plus consommateurs en énergie. L'utilisation des centres de données se partage entre le calcul intensif et les services web, aussi appelés informatique en nuage. La rapidité de calcul est primordiale pour le calcul intensif, mais pour les autres services ce paramètre peut varier selon les accords signés sur la qualité de service. Certains centres de données sont dits hybrides car ils combinent plusieurs types de services. Toutes ces infrastructures sont extrêmement énergivores. Dans ce présent manuscrit nous étudions les modèles de consommation énergétiques des systèmes informatiques. De tels modèles permettent une meilleure compréhension des serveurs informatiques et de leur façon de consommer l'énergie. Ils représentent donc un premier pas vers une meilleure gestion de ces systèmes, que ce soit pour faire des économies d'énergie ou pour facturer l'électricité à la charge des utilisateurs finaux. Les politiques de gestion et de contrôle de l'énergie comportent de nombreuses limites. En effet, la plupart des algorithmes d'ordonnancement sensibles à l'énergie utilisent des modèles de consommation restreints qui renferment un certain nombre de problèmes ouverts. De précédents travaux dans le domaine suggèrent d'utiliser les informations de contrôle fournies par le système informatique lui-même pour surveiller la consommation énergétique des applications. Néanmoins, ces modèles sont soit trop dépendants du type d'application, soit manquent de précision. Ce manuscrit présente des techniques permettant d'améliorer la précision des modèles de puissance en abordant des problèmes à plusieurs niveaux: depuis l'acquisition des mesures de puissance jusqu'à la définition d'une charge de travail générique permettant de créer un modèle lui aussi générique, c'est-à-dire qui pourra être utilisé pour des charges de travail hétérogènes. Pour atteindre un tel but, nous proposons d'utiliser des techniques d'apprentissage automatique.Les modèles d'apprentissage automatique sont facilement adaptables à l'architecture et sont le cœur de cette recherche. Ces travaux évaluent l'utilisation des réseaux de neurones artificiels et la régression linéaire comme technique d'apprentissage automatique pour faire de la modélisation statistique non linéaire. De tels modèles sont créés par une approche orientée données afin de pouvoir adapter les paramètres en fonction des informations collectées pendant l'exécution de charges de travail synthétiques. L'utilisation des techniques d'apprentissage automatique a pour but d'atteindre des estimateurs de très haute précision à la fois au niveau application et au niveau système. La méthodologie proposée est indépendante de l'architecture cible et peut facilement être reproductible quel que soit l'environnement. Les résultats montrent que l'utilisation de réseaux de neurones artificiels permet de créer des estimations très précises. Cependant, en raison de contraintes de modélisation, cette technique n'est pas applicable au niveau processus. Pour ce dernier, des modèles prédéfinis doivent être calibrés afin d'atteindre de bons résultats. / The number of computing systems is continuously increasing during the last years. The popularity of data centers turned them into one of the most power demanding facilities. The use of data centers is divided into high performance computing (HPC) and Internet services, or Clouds. Computing speed is crucial in HPC environments, while on Cloud systems it may vary according to their service-level agreements. Some data centers even propose hybrid environments, all of them are energy hungry. The present work is a study on power models for computing systems. These models allow a better understanding of the energy consumption of computers, and can be used as a first step towards better monitoring and management policies of such systems either to enhance their energy savings, or to account the energy to charge end-users. Energy management and control policies are subject to many limitations. Most energy-aware scheduling algorithms use restricted power models which have a number of open problems. Previous works in power modeling of computing systems proposed the use of system information to monitor the power consumption of applications. However, these models are either too specific for a given kind of application, or they lack of accuracy. This report presents techniques to enhance the accuracy of power models by tackling the issues since the measurements acquisition until the definition of a generic workload to enable the creation of a generic model, i.e. a model that can be used for heterogeneous workloads. To achieve such models, the use of machine learning techniques is proposed. Machine learning models are architecture adaptive and are used as the core of this research. More specifically, this work evaluates the use of artificial neural networks (ANN) and linear regression (LR) as machine learning techniques to perform non-linear statistical modeling.Such models are created through a data-driven approach, enabling adaptation of their parameters based on the information collected while running synthetic workloads. The use of machine learning techniques intends to achieve high accuracy application- and system-level estimators. The proposed methodology is architecture independent and can be easily reproduced in new environments.The results show that the use of artificial neural networks enables the creation of high accurate estimators. However, it cannot be applied at the process-level due to modeling constraints. For such case, predefined models can be calibrated to achieve fair results.% The use of process-level models enables the estimation of virtual machines' power consumption that can be used for Cloud provisioning.

This paper's main objective is to consolidate the knowledge on context in the realm of intelligent systems, systems that are aware of their context and can adapt their behavior accordingly. We provide an overview and analysis of 36 context models that are heterogeneous and scattered throughout multiple fields of research. In our analysis, we identify five shared context categories: social context, location, time, physical context, and user context. In addition, we compare the context models with the context elements considered in the discourse on intelligent systems and find that the models do not properly represent the identified set of 3,741 unique context elements. As a result, we propose a consolidation of the findings from the 36 context models and the 3,741 unique context elements. The analysis reveals that there is a long tail of context categories that are considered only sporadically in context models. However, particularly these context elements in the long tail may be necessary for improving intelligent systems' context awareness.

Driver distraction remains a matter of concern throughout the world as the number of car accidents caused by distracted driving is still unacceptably high. Industry and academia are working intensively to design new techniques that will address all types of driver distraction including visual, manual, auditory and cognitive distraction. This research focuses on an existing technology, namely in-car communication systems (ICCS). ICCS allow drivers to interact with their mobile phones without touching or looking at them. Previous research suggests that ICCS have reduced visual and manual distraction. Two problems were identified in this research: existing ICCS are still expensive and only available in limited models of car. As a result of that, only a small number of drivers can obtain a car equipped with an ICCS, especially in developing countries. The second problem is that existing ICCS are not aware of the driving context, which plays a role in distracting drivers. This research project was based on the following thesis statement: A mobile, context-aware model can be designed to reduce driver distraction caused by the use of ICCS. A mobile ICCS is portable and can be used in any car, addressing the first problem. Context-awareness will be used to detect possible situations that contribute to distracting drivers and the interaction with the mobile ICCS will be adapted so as to avert calls and text messages. This will address the second problem. As the driving context is dynamic, drivers may have to deal with critical safety-related tasks while they are using an existing ICCS. The following steps were taken in order to validate the thesis statement. An investigation was conducted into the causes and consequences of driver distraction. A review of literature was conducted on context-aware techniques that could potentially be used. The design of a model was proposed, called the Multimodal Interface for Mobile Info-communication with Context (MIMIC) and a preliminary usability evaluation was conducted in order to assess the feasibility of a speech-based, mobile ICCS. Despite some problems with the speech recognition, the results were satisfying and showed that the proposed model for mobile ICCS was feasible. Experiments were conducted in order to collect data to perform supervised learning to determine the driving context. The aim was to select the most effective machine learning techniques to determine the driving context. Decision tree and instance-based algorithms were found to be the best performing algorithms. Variables such as speed, acceleration and linear acceleration were found to be the most important variables according to an analysis of the decision tree. The initial MIMIC model was updated to include several adaptation effects and the resulting model was implemented as a prototype mobile application, called MIMIC-Prototype.

Master of Philosophy (Architecture) / Context-aware computing is a mobile computing paradigm in which applications can discover, use, and take advantage of contextual information, such as the location, tasks and preferences of the user, in order to adapt their behaviour in response to changing operating environments and user requirements. A problem that arises is the inability to respond to contextual information that cannot be classified into any known context. Many context-aware applications require all discovered contextual information to exactly match a type of context, otherwise the application will not react responsively. The ability to learn and recall contexts based on the contextual information discovered has not been very well addressed by previous context-aware applications and research. The aim of this thesis is to develop a component middleware technology for mobile computing devices for the discovery and capture of contextual information, using the situated reasoning concept of constructive memory. The research contribution of this thesis lies in developing a modified architecture for context-aware systems, using a constructive memory model as a way to learn and recall contexts from previous experiences and application interactions. Using a constructive memory model, previous experiences can be induced to construct potential contexts, given a small amount of learning and interaction. The learning process is able to map the many variations of contextual information currently discovered by the user with a predicted type of context based on what the application has stored and seen previously. It only requires a small amount of contextual information to predict a context, something common context-aware systems lack, as they require all information before a type of context is assigned. Additionally, some mechanism to reason about the contextual information being discovered from past application interactions will be beneficial to induce contexts for future experiences.

Master of Philosophy (Architecture) / Context-aware computing is a mobile computing paradigm in which applications can discover, use, and take advantage of contextual information, such as the location, tasks and preferences of the user, in order to adapt their behaviour in response to changing operating environments and user requirements. A problem that arises is the inability to respond to contextual information that cannot be classified into any known context. Many context-aware applications require all discovered contextual information to exactly match a type of context, otherwise the application will not react responsively. The ability to learn and recall contexts based on the contextual information discovered has not been very well addressed by previous context-aware applications and research. The aim of this thesis is to develop a component middleware technology for mobile computing devices for the discovery and capture of contextual information, using the situated reasoning concept of constructive memory. The research contribution of this thesis lies in developing a modified architecture for context-aware systems, using a constructive memory model as a way to learn and recall contexts from previous experiences and application interactions. Using a constructive memory model, previous experiences can be induced to construct potential contexts, given a small amount of learning and interaction. The learning process is able to map the many variations of contextual information currently discovered by the user with a predicted type of context based on what the application has stored and seen previously. It only requires a small amount of contextual information to predict a context, something common context-aware systems lack, as they require all information before a type of context is assigned. Additionally, some mechanism to reason about the contextual information being discovered from past application interactions will be beneficial to induce contexts for future experiences.

Context-aware computing aims at making mobile devices sensitive to the social and physical settings in which they are used. A necessary requirement to achieve this goal is to enable those devices to establish a shared understanding of the desired settings. Establishing a shared understanding entails the need to manipulate sensed data in order to capture a real world situation wholly, conceptually, and meaningfully. Quite often, however, the data acquired from sensors can be inexact, incomplete, and/or uncertain. Inexact sensing arises mostly due to the inherent limitation of sensors to capture a real world phenomenon precisely. Incompleteness is caused by the absence of a mechanism to capture certain real-world aspects; and uncertainty stems from the lack of knowledge about the reliability of the sensing sources, such as their sensing range, accuracy, and resolution. The thesis identifies a set of criteria for a context-aware system to capture dynamic real-world situations. On the basis of these criteria, a distributed architecture is designed, implemented and tested. The architecture consists of Primitive Context Servers, which abstract the acquisition of primitive contexts from physical sensors; Aggregators, to minimise error caused by inconsistent sensing, and to gather correlated primitive contexts pertaining to a particular entity or situation; a Knowledge Base and an Empirical Ambient Knowledge Component, to model dynamic properties of entities with facts and beliefs; and a Composer, to reason about dynamic real-world situations on the basis of sensed data. Two additional components, namely, the Event Handler and the Rule Organiser, are responsible for dynamically generating context rules by associating decision events ? signifying a user?s activity ? with the context in which those decision events are produced. Context-rules are essential elements with which the behaviour of mobile devices can be controlled and useful services can be provided. Four estimation and recognition schemes, namely, Fuzzy Logic, Hidden Markov Models, Dempster-Schafer Theory of Evidence, and Bayesian Networks, are investigated, and their suitability for the implementation of the components of the architecture of the thesis is studied. Subsequently, fuzzy sets are chosen to model dynamic properties of entities. Dempster-Schafer?s combination theory is chosen for aggregating primitive contexts; and Bayesian Networks are chosen to reason about a higher-level context, which is an abstraction of a real-world situation. A Bayesian Composer is implemented to demonstrate the capability of the architecture in dealing with uncertainty, in revising the belief of the Empirical Ambient Knowledge Component, in dealing with the dynamics of primitive contexts and in dynamically defining contextual states. The Composer could be able to reason about the whereabouts of a person in the absence of any localisation sensor. Thermal, relative humidity, light intensity properties of a place as well as time information were employed to model and reason about a place. Consequently, depending on the variety and reliability of the sensors employed, the Composer could be able to discriminate between rooms, corridors, a building, or an outdoor place with different degrees of uncertainty. The Context-Aware E-Pad (CAEP) application is designed and implemented to demonstrate how applications can employ a higher-level context without the need to directly deal with its composition, and how a context rule can be generated by associating the activities (decision events) of a mobile user with the context in which the decision events are produced.

Context-aware computing aims at making mobile devices sensitive to the social and physical settings in which they are used. A necessary requirement to achieve this goal is to enable those devices to establish a shared understanding of the desired settings. Establishing a shared understanding entails the need to manipulate sensed data in order to capture a real world situation wholly, conceptually, and meaningfully. Quite often, however, the data acquired from sensors can be inexact, incomplete, and/or uncertain. Inexact sensing arises mostly due to the inherent limitation of sensors to capture a real world phenomenon precisely. Incompleteness is caused by the absence of a mechanism to capture certain real-world aspects; and uncertainty stems from the lack of knowledge about the reliability of the sensing sources, such as their sensing range, accuracy, and resolution. The thesis identifies a set of criteria for a context-aware system to capture dynamic real-world situations. On the basis of these criteria, a distributed architecture is designed, implemented and tested. The architecture consists of Primitive Context Servers, which abstract the acquisition of primitive contexts from physical sensors; Aggregators, to minimise error caused by inconsistent sensing, and to gather correlated primitive contexts pertaining to a particular entity or situation; a Knowledge Base and an Empirical Ambient Knowledge Component, to model dynamic properties of entities with facts and beliefs; and a Composer, to reason about dynamic real-world situations on the basis of sensed data. Two additional components, namely, the Event Handler and the Rule Organiser, are responsible for dynamically generating context rules by associating decision events ? signifying a user?s activity ? with the context in which those decision events are produced. Context-rules are essential elements with which the behaviour of mobile devices can be controlled and useful services can be provided. Four estimation and recognition schemes, namely, Fuzzy Logic, Hidden Markov Models, Dempster-Schafer Theory of Evidence, and Bayesian Networks, are investigated, and their suitability for the implementation of the components of the architecture of the thesis is studied. Subsequently, fuzzy sets are chosen to model dynamic properties of entities. Dempster-Schafer?s combination theory is chosen for aggregating primitive contexts; and Bayesian Networks are chosen to reason about a higher-level context, which is an abstraction of a real-world situation. A Bayesian Composer is implemented to demonstrate the capability of the architecture in dealing with uncertainty, in revising the belief of the Empirical Ambient Knowledge Component, in dealing with the dynamics of primitive contexts and in dynamically defining contextual states. The Composer could be able to reason about the whereabouts of a person in the absence of any localisation sensor. Thermal, relative humidity, light intensity properties of a place as well as time information were employed to model and reason about a place. Consequently, depending on the variety and reliability of the sensors employed, the Composer could be able to discriminate between rooms, corridors, a building, or an outdoor place with different degrees of uncertainty. The Context-Aware E-Pad (CAEP) application is designed and implemented to demonstrate how applications can employ a higher-level context without the need to directly deal with its composition, and how a context rule can be generated by associating the activities (decision events) of a mobile user with the context in which the decision events are produced.