Search

Was Bits & Bäume verbindet: Digitalisierung nachhaltig gestalten, ISBN:978-3-96238-149-3

A large fraction of energy consumed in households is due to space heating. Especially during daytime, the heating is often running constantly, controlled only by a thermostat – even if the inhabitants are not present. Taking advantage of the absence of the inhabitants to save heating energy by lowering the temperature thus poses a great opportunity. Since the concrete savings of an occupancy-based heating strategy strongly depend on the individual occupancy pattern, a fast and inexpensive method to quantify these potential savings would be beneficial. In this paper we present such a practical method which builds upon an approach to estimate a household’s occupancy from its historical electricity consumption data, as gathered by smart meters. Based on the derived occupancy data, we automatically calculate the potential savings. Besides occupancy data, the underlying model also takes into account publicly available weather data and relevant building characteristics. Using this approach, households with high potential for energy savings can be quickly identified and their members could be more easily convinced to adopt an occupancy-based heating strategy (either by manually adjusting the thermostat or by investing in automation) since their monetary benefits can be calculated and the risk of misinvestment is thus reduced. To prove the usefulness of our system, we apply it to a large dataset containing relevant building and household data such as the size and age of several thousand households and show that, on average, a household can save over 9% heating energy when following an occupancy-based heating regime, while certain groups, such as single-person households, can even save 14% on average.

This report presents a simulation framework based on the ISO 13790 5R1C lumped capacitance building model for the evaluation of the energy savings potential of occupancy prediction algorithms. We show the derivation of the model parameters and introduce a new methodology to prepare weather data for simulating the energy consumption of a heating system when predictively controlling the thermostat.

Increasing number of appliances drives residential energy consumption

When it comes to conserving electricity, it is crucial for users to know how much energy is consumed by individual appliances. However, the technical feedback provided by existing energy consumption feedback systems in the form of dry numbers and intangible units is not appropriate for most users. To address this shortcoming we developed PowerPedia, a system that provides behavior-influencing feedback over and above pure consumption values. By integrating a community platform - a Wikipedia for electrical appliances - PowerPedia enables users to identify and compare the consumption of their domestic appliances with that of others. It thus helps users to better understand their electricity consumption and take effective action to save electricity

Technischer Bericht / Departement Informatik, ETH Zürich, 507

In this paper we introduce a new distributed algorithm for location discovery. It can be used in wireless ad-hoc sensor networks that are equipped with means of measuring the distances between the nodes (like the intensity of the received signal strength). The algorithm takes the reliability of measurements into account during calculation of the nodes positions. Simulation results are presented, showing the algorithms performance in relation to its accuracy, communication and calculation costs. The simulation results of our approach yield 2 to 4 times better results in position accuracy than other systems described previously. This level of performance can be reached using only few broadcast messages with small and constant size, for each node in the network.

Arbeitsbericht

The paper shows that characterizing the causal relationship between significant events is an important but non-trivial aspect for understanding the behavior of distributed programs. An introduction to the notion of causality and its relation to logical time is given; some fundamental results concerning the characterization of causality are pre- sented. Recent work on the detection of causal relationships in distributed computations is surveyed. The relative merits and limitations of the different approaches are discussed, and their general feasibility is analyzed.

The paper presents several algorithmic solutions to typical problems from the theory of distributed computing. The following topics are treated: Distributed approximation, leader election, routing tables for shortest paths, termination detection, parallel graph traversal, information dissemination, consistent snapshot computation. Concepts like atomic actions, message driven computations, time diagrams, and consistent cuts are introduced and references to the literature for further reading are given.

Over the last decades, embedded computer systems have become powerful and wide-spread with remarkable success. Besides traditional computers, such as desktops, laptops, smartphones, and servers, such systems have become part of nearly every technical appliance, for example, cars, televisions, and washing machines and thereby, an essential part of our lives. A common phrase for such appliances is “smart device”, a term which encompasses equipment to which one can digitally connect in order to exchange information and commands. Further, it can potentially sense its environment and process as well as act upon this measurement. Although computers and the devices containing them play such an important role, the way in which we interact with them has not changed much since the early days. Humans are required to control them in a manner very different from human-to-human communication. In particular, the possibilities to provide input to a smart device are mostly limited to traditional interfaces, such as buttons, knobs, or keyboards; or graphical representations thereof on displays. Technological progress in recent years in hardware, as well as in algorithmic methods, e.g. in machine learning, enables novel solutions for human-computer (or in fact human-device) interaction. Only recently, the use of speech has become practical and is used on smartphones as well as for home devices, incorporating a modality in the interaction process that is innate to humans and rich in expressiveness. However, for natural communication, other modalities, such as gestures, complement speech and may do so in human-computer communication, enabling simple and spontaneous interactions and avoiding the known social awkwardness of having to talk to devices. The adoption of speech and also other commonly used interaction methods, such as touch input on smartphones, indicate the relevance of considering further modalities in addition to the traditional ones. This dissertation contributes to further bridging the interaction gap between humans and smart devices by exploring solutions in the following areas: (i) Wearable gesture recognition based on electromyography (EMG): The touch of our fingers is widely used for interaction. However, most approaches only consider binary touch events. We present a method, which classifies finger touches using a novel neural network architecture and estimates their force based on data recorded from a wireless EMG armband. Our method runs in real time on a smartphone and allows for new interactions with devices and objects, as any surface can be turned into an interactive surface and additional functionality can be encoded through single fingers and the force applied. (ii) Wearable gesture recognition based on sound and motion: Besides other signals, gestures might also emit sound. We develop a recognition method for sound-emitting gestures, such as snapping, knocking, or clapping, employing only a standard smartwatch. Besides the motion information from the built-in accelerometer and gyroscope, we exploit audio data recorded by the smartwatch microphone as input. We propose a lightweight convolutional neural network architecture for gesture recognition, specifically designed to run locally on resource-constrained devices. It achieves a user-independent recognition accuracy of 97.2% for nine distinct gestures. We find that the audio input drastically reduces the false positive rate in continuous recognition compared to using only motion. (iii) Device representations in wearable augmented reality (AR): While AR technology is becoming increasingly available to the public, ways of interaction in the AR space are not yet fully understood. We investigate how users can control smart devices in AR. Connected devices are augmented with interaction widgets representing them. For example, a widget can be overlaid on a loudspeaker to control its volume. We explore three ways of manipulating the virtual widgets in a user study: (1) in-air finger pinching and sliding, (2) whole-arm gestures rotating and waving, (3) incorporating physical objects in the surroundings and mapping their movements to interaction primitives. We find significant differences in the preference of the users, the speed of executing commands, and the granularity of the type of control. While these methods only apply to control of a single device at a time, in a second step, we create a method which also takes potential connections between devices into account. Users can view, create, and manipulate connections between smart devices in AR using simple gestures. (iv) Personalizable user interfaces from simple materials: User interfaces rarely adapt to specific user preferences or the task at hand. We present a method that allows the quick and inexpensive creation of personalized interfaces from paper. Users can cut out shapes and assign control functions to these paper snippets via a simple configuration interface. After configuration, control takes place entirely through the manipulation of the paper shapes, providing the experience of a tailored tangible user interface. The shapes, which are monitored by a camera with depth sensing, can be dynamically changed during use. The proposed methods aim at a more natural interaction with smart devices through advanced sensing and processing in the user’s environment or on his/her body itself. As these interactions could be made ubiquitously available through wearable computers, our methods could help to improve the usability of the growing number of smart devices and make them more easily accessible to more people.

With an ever-increasing number of smart devices in the users' surrounding that create ubiquitous opportunities for interaction, enabling natural and intuitive interaction with digital systems is a key challenge in Human-Computer Interaction (HCI). Computing technology is steadily getting smaller, cheaper, and more interconnected, including sensing and communication technology, currently witnessed by mobile and wearable devices of various forms. Smartwatches and fitness trackers are already ubiquitous, while smart glasses or augmented reality headsets such as the Microsoft HoloLens are closer to real products than to lab prototypes. Given such a changing landscape of devices in our environment, current input modalities may limit the effective utilisation of the available information flow. When people interact with each other, they use multiple modalities including voice, body posture, hand gestures, facial expressions, or eye gaze. Touch interactions coupled with apps on our smartphones are today’s universal interaction devices and have recently been extended with voice control. However, such input modalities are no longer sufficient for complex applications like augmented or virtual reality. In this dissertation, we broaden and facilitate the interface between technology and people and contribute to the following research areas: i) Collocated Multi-user Gestural Interactions. Natural gestures are a promising input modality in HCI because they enrich the way we interact with complex systems. However, few works have explored this input technique when multiple users are physically collocated. We explore multi-user interaction through gestures between people who are physically close to one another. The proximity of users is detected through inaudible acoustic ranging while in-air hand gestures are recognised by leveraging inertial sensors. To ensure scalability, the underlying communication protocol between users and devices is handled over Bluetooth. Through extensive evaluations, we demonstrate not only the robustness of our approach but also the feasibility of using off-the-shelf mobile and wearable devices. We first demonstrate our concept with HandshakAR, a wearable system that facilitates the exchange of digital information between two users. We then extend this concept to multiple users and show three different practical application scenarios. ii) Eye-based Human-Computer Interaction. We investigate eye gaze as a hands-free high-bandwidth input modality and explore how users can augment their interaction capabilities with the gaze direction. We propose ubiGaze, a novel wearable system that enables attaching virtual content to any real-world object through gaze gestures, which are detected using a wearable eye tracker. While gaze gestures are less sensitive to accuracy problems or calibration shifts, many practical applications require calibrated eye trackers. However, existing calibration techniques are tedious and rely on special markers. We propose finger calibration, a novel method for head-mounted eye trackers in which users only have to point with their fingers at locations in the scene. This eliminates the need for additional assistance or specialised markers. Lastly, while eye trackers have become more accessible, the need for special-purpose equipment hinders many large-scale deployments. Therefore, we explore gaze-based interaction using a single off-the-shelf camera. We propose a method to detect pursuit eye movements, which have become widely popular because they enable spontaneous interaction. Our method combines appearance-based gaze estimation with optical flow in the eye region to jointly analyse eye movement dynamics in a single pipeline. Our results not only show the feasibility of our approach but point the way towards new methods that only require standard cameras, which are readily available in an ever increasing number of devices. iii) Quantification of Visual Attention in Mobile HCI. Eye contact is a key measure of overt visual attention in mobile HCI as it enables understanding when, how often, or for how long users look at their devices. However, robustly detecting shifts of attention during everyday mobile interactions is challenging. Encouraged by recent advances in automatic eye contact detection, we provide a fundamental investigation into the feasibility of quantifying mobile visual attention. We identify core challenges and sources of errors associated with sensing visual attention in the wild, including the impact of face and eye visibility, the importance of robust head pose estimation, and the need for accurate gaze estimation. Guided by our analysis, we present a method to accurately and robustly detect eye contact in images captured with the front-facing camera of common mobile devices. Based on our evaluations, we show how eye contact is the fundamental building block for calculating higher-level attention metrics and, as such, enables studying visual attention in the wild. Finally, we present the Everyday Mobile Visual Attention (EMVA) dataset and quantitative evaluations of visual attention of mobile device users in-situ, i.e. while they use their devices during everyday routine. Using our proposed method for eye contact detection, we quantify the highly dynamic nature of everyday visual attention allocation across users, mobile applications, and usage contexts. We then discuss our findings which highlight the potential and inform the design of future mobile attentive user interfaces. Along with this dissertation, we deliver open-source implementations of some of our contributions. The fingertip calibration method is available as a plugin for the open-source Pupil platform. Additionally, the EMVA dataset, which contains around 472 hours of video snippets from 32 participants recorded over more than two weeks in real life using the front-facing camera as well as associated usage logs, interaction events, and sensor data, is publicly available to support future work in this area of research.

As Internet of Things (IoT) systems further emerge, we face unprecedented security and privacy challenges, especially with regards to the collected data. This data typically consists of sensor readings, tagged with metadata. For scalability, ubiquitous access, and sharing possibilities, the data is most often stored in the cloud. Securing date while in transit and in particular when being stored in the cloud is of utmost importance, as the data can be used to infer privacy-sensitive information. Moreover, transparent and secure data sharing (e.g., sharing with friends or domain experts) is considered a key requirement for the practicality and success of typical IoT systems. In today’s cloud-centric designs, users have no choice but to trust centralized parties. The increased number of security and privacy incidents, such as system compromises or unauthorized trade with users data, show that this trust is not always justified. Despite varying levels of privacy-awareness among users of different age and geopolitical groups, and even societal shifts towards privacy pragmatism and indifference, the security and privacy threats do usually have far-reaching implications, demanding adequate mechanisms and measures to address them. In this dissertation, we investigate building secure IoT systems that protect data confidentiality and retain data ownership. We build secure systems that allow reducing the trust end-users are required to put into third parties within the IoT ecosystem, specifically towards the cloud storage and service providers. More importantly, we take a new approach on empowering the user with ownership and fine-grained access control for IoT data without sacrificing performance or security. In particular, we present three approaches to enabling a secure IoT ecosystem: (i) Talos: Talos is a system that stores IoT data securely in a cloud database while still allowing query processing over the encrypted data. Talos protects data even if the server is compromised. We enable this by encrypting IoT data with a set of cryptographic schemes such as order- preserving and partially homomorphic encryption. We tailor Talos to accommodate for the resource asymmetry of the IoT, particularly towards constrained IoT devices. We assess the feasibility of Talos on low-power devices with and without cryptographic hardware accelerators and quantify its overhead concerning energy consumption, computation time, and latency. With a thorough evaluation of our prototype implementation, we show that Talos is a practical system that can provide a high level of security with reasonable overhead. (ii) Pilatus: Storage of data on cloud services naturally facilitates data sharing with third-party services and other users, but bears privacy risks. We present Pilatus, a data protection platform that extends Talos where the cloud stores only encrypted data, yet is still able to process a defined set of database queries (e.g., range or sum). Pilatus features a novel encrypted data sharing scheme based on re-encryption, with revocation capabilities and in situ key-update. Our solution includes a suite of novel techniques that enable efficient partially homomorphic encryption, decryption, and sharing. We present performance optimizations that render these cryptographic tools practical for mobile platforms. We implement a prototype of Pilatus and evaluate it thoroughly. Our optimizations achieve a performance gain within one order of magnitude compared to state-of-the-art realizations. (iii) Droplet: Droplet is a secure data management system that we designed from the ground up to accommodate for the distributed nature of the IoT and revive the IoT from the current vertical design paradigm. The consequent myriad of isolated data silos of classical vertical architectures is hard to manage and prevent heterogeneous applications from interacting with our IoT data. To address this challenge, we leverage the blockchain technology to bootstrap trust for a distributed, secure, and resilient access control and data management scheme. Droplet handles time series data, enables reliable sharing among heterogeneous applications without intermediate trust entities, and features a cryptographically-protected fine-grained and scalable access control mechanism to data streams. We leverage a hash-chain-based key management mechanism to enable interval sharing and compact key distribution. The built-in cryptocurrency feature of blockchains allows the integration of economic incentives into our system. These properties enable a variety of applications that are presently not easily realizable using existing systems. The systems proposed and discussed in this dissertation demonstrate that end-to-end encryption with secure sharing can be achieved in IoT ecosystems with a modest overhead, while maintaining a consistent user- experience.

Aiming for sustainable development means reconsidering the access to energy sources in industrialized countries, which are not faced with contingency scenarios that are implemented in emergent and newly-developed countries, to allow equal access to energy sources for all and thwart environmental degradation. The global penetration of renewable energy sources to replace fossil fuel and nuclear power plants means adjusting to stochastic energy production. The expected yield will be dependent on very different weather and landscape conditions and will represent a challenge for countries with continuous access to energy sources and where energy is often considered a public utility. Tracking wastage, improving the scheduling, and the processes that consume energy, would allow us to match the demand and the supply of energy. This will be particularly crucial during peak time, where meeting the high demand incurs the ramping up of mostly unclean additional power plants or introducing power system instability. The digitalization of the energy sector has started with the roll- out of smart meters to record the electricity consumption at a finer granularity and are aimed to replace the biannual or yearly dispatch of utility companies’ employees to read the meter. Considerable research efforts have been directed at analyzing aggregated loads from these smart meters or at developing algorithms for disaggregating households’ total electricity consumption to isolate single appliances’ traces. However, less focus has been set on assessing the potential of using sub-metered data for improving the energy efficiency in households. This was primarily linked to the fact that the necessary datasets were not widely available, due to the difficulty and the costs in instrumenting households for acquiring the consumption data from appliances. The objective of this thesis is to investigate how to leverage and improve existing disaggregated datasets to develop data-driven techniques to improve the energy efficiency within residential homes. Starting from smart meter data, we segmented households into groups with similar electricity consumption pattern based on their peak consumption, to identify hurtful consumption patterns in the perspective of utility companies, for which they could launch targeted mitigation campaigns. However, improving the energy efficiency in the residential sector requires to change individuals’ relationship to- wards their electricity consumption. These behaviors are closely re- lated to the activities that are carried out throughout the day and can be supported by the usage of consumer electronics, such as appliances. Therefore, we turned to analyzing the behaviors inside house- i holds that triggered the usage of electricity by studying a large disaggregated dataset and developed learning techniques to extract activity patterns. We first addressed the challenge of determining when appliances are actively used by households’ residents, from when they are off or idle and incurring standby consumption by developing GMMthresh, an automatic thresholding method, which is agnostic of the appliance’s type, brand and model, but instead relies on the statistical distribution of its power consumption. Due to the lack of event-based and activity labels in existing datasets to allow us to validate our learning technique, we leveraged crowdsourcing concepts to provide an expert-annotated dataset to enrich the existing datasets through our Collaborative Annotation Framework for Energy Datasets (CAFED). We conducted two in-depth studies to quantify the performance of regular users against expert users in labeling energy data on CAFED. We provided analysis tools and methods that can be generalized to crowdsourcing systems for improving the quality of the workers’ contributions. Using the expert-annotated labels, we validated GMMthresh with expert manually labeled data. Then, we developed a method for learning temporal association rules for identifying activities involving the usage of appliances within households. Our pipeline includes our thresholding algorithm and a novel search algorithm for determining time windows for the association rules efficiently and in a data-driven manner. The contributions of this thesis rely on exploiting energy data and developing novel techniques towards identifying activity patterns and their scheduling, which could then become part of an ambient intelligence system that would smarten existing homes. The methods we developed are not restricted to the energy research, as they can be applied to sensor data, where for example inertial sensors also require machine learning algorithms to filter out background noise from actual movement. Similarly, our work on the crowdsourcing of time series opens new perspectives for extending the range of data that can be annotated by the crowd and provides design insights and mitigation techniques for improving the quality of the labeling on collaborative platforms. Finally, our temporal association rules mining framework is not limited to energy time series but can be applied to search for temporal windows and understanding the scheduling of any time series dataset.

We extend Clarkson's framework by considering parameterized convex optimization problems over the unit simplex, that depend on one parameter. We provide a simple and efficient scheme for maintaining an e-approximate solution (and a corresponding e-coreset) along the entire parameter path. We prove correctness and optimality of the method. Practically relevant instances of the abstract parameterized optimization problem are for example regularization paths of support vector machines, multiple kernel learning, and minimum enclosing balls of moving points.

Zeitreihen kommen unter anderem in vielzähligen Bereichen der Medizin, Astronomie, Geophysik, Konstruktion, und Finanzwirtschaft vor. Im Allgemeinen bezeichnet man eine Zeitreihe als Sequenz von Datenpunkten die mit fortlaufender Zeit in gleichmäßigen Zeitabständen gemessen wurde. Diese These beschäftigt sich hauptsächlich mit der Auswertung von Zeitreihen, was die Segmentierung, Klassifikation, und Gruppierung von temporalen Daten beinhaltet. Viele Algorithmen die diese Aufgaben lösen bedingen den paarweisen Vergleich von Sequenzen, was das fortwährende Forschungsinteresse an Distanzmaßen als entscheidende Subroutine begründet. Im Verlauf dieser Arbeit führen wir mehrere neue Distanzmaße ein welche die wesentlichen Charakteristiken von Zeitreihen erfassen und die Unterscheidung von verschiedenen, in einem Datensatz vorkommenden, Klassen ermöglichen. Unsere vorgeschlagenen Distanzmaße adressieren häufige, bei der Auswertung von Zeitreichen auftretende, Herausforderung. Dazu gehören die Untersuchung von multivariaten Daten, der Rechenaufwand von paarweisen Ähnlichkeitsberechnungen, die Messstörungen und Verzerrungen von temporalen Daten, das Trennen von gemischten Signalen, sowie die Analyse von nicht linearen Systemen. Unsere Arbeit leistet einen Betrag im Gebiet der Zeitreihenanalyse indem wir neue Ansätze zur Erkennung von Mustern in temporalen Daten einführen, robuste Distanzmaße für die effiziente Klassifikation von Zeitreihen bereitstellen, zuvor unbeachtete Invarianz betrachten und entsprechende Distanzberechnungen vorschlagen, unterschiedliche Methoden des Maschinellen Lernens für die Trennung von Signalen vergleichen, und nicht lineare Model für die Untersuchung von Zeitreichen adaptieren. Des Weiteren demonstrieren wir die Einsetzbarkeit unserer vorgeschlagen Distanzmaße in praktischen Anwendungen, wie z.B. bei der Optimierung von Fahrzeugmotoren in Bezug auf den Schadstoffausstoß sowie die Optimierung von Heizplänen unter Betrachtung des Energieverbrauches. Darüber hinaus präsentieren wir mehrere eigens entwickelte Zeitreihenanalysewerkzeuge die unsere eingeführten Distanzmaße anwenden. Time series can be found in domains as diverse as medicine, astronomy, geophysics, engineering, and quantitative finance. In general, a time series is a sequence of data points, measured at successive points in time and spaced at uniform time intervals. This thesis is concerned with time series mining, including segmentation, classification, and clustering of temporal data. Many algorithms for these tasks depend upon pairwise (dis)similarity comparisons of (sub)sequences, which accounts for the continued research on time series distance measures as an important subroutine. In the course of this work we introduce several novel distance measures, which describe time series characteristics that may distinguish the individual classes contained in the data. Our proposed time series distance measures address frequently encountered issues, such as the processing of multivariate data, the computational complexity of pairwise (dis)similarity comparisons, the invariance required for temporal data with distortions, the separation of mixed signals, and the analysis of nonlinear systems. Our work contributes to the time series community by introducing novel approaches to pattern recognition in temporal data, presenting miscellaneous sensor fusion techniques for multivariate measurements, offering efficient and robust distance measures for fast time series classification, introducing previously disregarded invariance and proposing corresponding distance measures, comparing various machine learning algorithms for signal separation, and providing nonlinear models for time series mining. In addition to our theoretical contributions, we furthermore demonstrate that our proposed time series distance measures are beneficial in real-world applications, including the optimization of vehicle engines with regard to exhaust emission and the optimization of heating control in terms of energy efficiency. Furthermore, we present several specifically developed time series mining tools, which implement our introduced distance measures and provide graphical user interfaces for straightforward parameter setting as well as exploratory data analysis.

A fundamental paradigm shift is currently taking place in the field of computing: due to the miniaturization of computing devices and the proliferation of embedded systems, tiny, networked computers can now be easily integrated into everyday objects, turning them into smart things. In the resulting Internet of Things, physical items are no longer disconnected from the virtual world but rather become accessible through computers and other networked devices, and can even make use of protocols that are widely deployed in the World Wide Web, in a paradigm that we call the Web of Things. Eventually, smart things will be able to communicate, analyze, decide, and act - and thereby provide an invisible background assistance that should make life more enjoyable, entertaining, and also safer. However, in an environment that is populated by hundreds of Web-enabled smart things, it will become increasingly difficult for humans to interact with devices that are relevant to their current needs, and to find, select, and control them. The objective of this thesis is to investigate how human users could be enabled to conveniently interact with individual smart objects in their surroundings and to interconnect devices and configure the resulting physical mashups to perform higher-level tasks on their behalf. To achieve basic interoperability between devices, we rely on the World Wide Web with its proven protocols and architectural patterns which emphasize scalability, generic interfaces, and loose coupling between components. As a first step to facilitate the interaction with smart things on top of the basic Web principles, we propose the embedding of metadata for automatically generating user interfaces for smart devices. Our specific approach enables not only the generation of more intuitive graphical widgets but also the mapping of interactive components to gesture-based, speech-based, and physical interfaces by describing the high-level interaction semantics of smart devices instead of specifying purely interface-specific information. The provisioning of an interaction mechanism with a smart object is thus reduced to the embedding of simple interaction information into the representation of the smart thing. Before users can start interacting with a smart device, it must, however, first be selected. To permit users to choose which of the many smart objects in their surroundings should be involved in an interaction, we propose to use technologies for optical image recognition. The visual selection of smart things and automatically generated user interfaces enable end users to conveniently interact with individual services in their surroundings that are embodied as specific physical objects. To complement the direct interaction with smart devices, the second part of this thesis focuses on more complex use cases where multiple smart objects must collaborate to achieve the user's goal. Such situations arise, for instance, in home or office automation scenarios, or in smart factories, where machines or assembly lines could adjust to better support the operator. To put users more in control of entire environments of smart devices, we present a system that records interactions between smart things and with remote services and displays this data to users in real time. To do this, we use an augmented reality overlay on the camera feed of handheld or wearable devices such as smartphones and smartglasses. Next, we propose a management infrastructure for smart things that makes the services they offer discoverable and composeable, and fully integrates them with more traditional Web-based information providers. This system enables humans to find and use data and functionality provided by physical devices and allows machines to support users in finding services within densely populated smart environments and even to discover and use required services themselves, on behalf of the user. The basis for these applications is a generic mechanism that allows smart devices to provide semantic descriptions of the services they offer. Specifically, our infrastructure supports the embedding of functional semantic metadata into smart things that describes which functionality a concrete object provides and how to invoke it. Based on this metadata, a semantic reasoning component can find out which composite tasks can be achieved by a user's smart environment and can provide instructions about how to reach concrete goals, thus enabling the configuration of entire smart environments for end users. As a concrete use case, we present a platform that applies our proposed interaction modes with smart things to automobiles: a mobile application recognizes cars, downloads information about them from a back-end server, and displays this information - as well as interaction capabilities with the car and its services - on the user's interface device. The back-end server furthermore exposes functional metadata about the capabilities of individual cars to make their services automatically usable within physical mashups. Finally, it records client interactions to enable car owners to monitor in real time who accesses which kind of data and services on their vehicles. The overarching objective of this thesis is to show how current technologies could support the interaction of end users with Web-enabled smart devices. To achieve this, we make use of a number of technologies from different areas of the computer science discipline: A management infrastructure makes smart things discoverable for human users and machines and builds upon current research in the distributed systems domain. State-of-the-art computer vision technologies allow users to select devices in their environment using handheld or wearable computers such as smartphones or smartglasses. Novel methods from the field of computer-human-interaction enable the embedding of metadata that allows for automatically generating user interfaces. Finally, semantic technologies enable flexible compositions of smart things that collaborate to achieve the user's goal.

Monitoring of health parameters in non-clinical settings is one strategy to address the increasingly aging population and age-related disabilities and diseases. However, challenges exist when introducing self-monitoring activities in people's everyday life. An active lifestyle can challenge the appropriation of healthcare technologies and people with comorbidity may have diverse but co-existing monitoring needs. In this paper, we seek to understand home-based health monitoring practices to better design and integrate them into people's everyday life. We perform an analysis of socio-technical complexities in home-based healthcare technologies through three case studies of self-monitoring: 1) pre-eclampsia (i.e. pregnancy poisoning), 2) heart conditions, and 3) preventive care. Through the analysis seven themes emerged (people, resources, places, routines, knowledge, control and motivation) that can facilitate the understanding of home-based healthcare activities. We present three modes of self-monitoring use and provide a set of design recommendations for future Ubicomp designs of home-based healthcare technology. Monitoring of health parameters in non-clinical settings is one strategy to address the increasingly aging population and age-related disabilities and diseases. However, challenges exist when introducing self-monitoring activities in people’s everyday life. An active lifestyle can challenge the appropriation of healthcare technologies and people with comorbidity may have diverse but co-existing monitoring needs. In this paper, we seek to understand home-based health monitoring practices to better design and integrate them into people’s everyday life. We perform an analysis of socio-technical complexities in home-based healthcare technologies through three case studies of self-monitoring: 1) pre-eclampsia (i.e. pregnancy poisoning), 2) heart conditions, and 3) preventive care. Through the analysis seven themes emerged (people, resources, places, routines, knowledge, control and motivation) that can facilitate the understanding of home-based healthcare activities. We present three modes of self-monitoring use and provide a set of design recommendations for future Ubicomp designs of home-based healthcare technology.

In this paper we present Local Area Artworks, a system enabling collaborative art interpretation on-site deployed during an exhibition in a local art gallery. Through the system, we explore ways to re-connect people to the local place by making use of their personal mobile devices as interfaces to the shared physical space. We do this by re-emphasizing the local characteristics of wireless networks over the global connectivity to the Internet. With a collaborative writing system in a semi-public place, we encourage local art discussions and provide a platform for the public to actively participate in interpretations of individual artworks. Our preliminary findings suggest that people were (to our surprise) not questioning the inner workings of our system. Through engaging with the system, individuals felt being part of the exhibition. However, no coherent piece of text emerged during the runtime of the exhibition. In this paper we present Local Area Artworks, a system enabling collaborative art interpretation on-site deployed during an exhibition in a local art gallery. Through the system, we explore ways to re-connect people to the local place by making use of their personal mobile devices as interfaces to the shared physical space. We do this by re-emphasizing the local characteristics of wireless networks over the global connectivity to the Internet. With a collaborative writing system in a semi-public place, we encourage local art discussions and provide a platform for the public to actively participate in interpretations of individual artworks. Our preliminary findings suggest that people were (to our surprise) not questioning the inner workings of our system. Through engaging with the system, individuals felt being part of the exhibition. However, no coherent piece of text emerged during the runtime of the exhibition.

Gathering representative data using mobile sensing to answer research questions is becoming increasingly popular, driven by growing ubiquity and sensing capabilities of mobile devices. However, there are pitfalls along this path, which introduce heterogeneity in the gathered data, and which are rooted in the diversity of the involved device platforms, hardware, software versions and participants. Thus, we, as a research community, need to establish good practices and methodologies for addressing this issue in order to help ensure that, e.g., scientific results and policy changes based on collective, mobile sensed data are valid. In this paper, we aim to inform researchers and developers about mobile sensing data heterogeneity and ways to combat it. We do so via distilling a vocabulary of underlying causes, and via describing their effects on mobile sensing---building on experiences from three projects within citizen science, crowd awareness and trajectory tracking.

We devise a framework for computing an approximate solution path for an important class of parameterized semidefinite problems that is guaranteed to be e-close to the exact solution path. The problem of computing the entire regularization path for matrix factorization problems such as maximum-margin matrix factorization fits into this framework, as well as many other nuclear norm regularized convex optimization problems from machine learning. We show that the combinatorial complexity of the approximate path is independent of the size of the matrix. Furthermore, the whole solution path can be computed in near linear time in the size of the input matrix. The framework employs an approximative semidefinite program solver for a fixed parameter value. Here we use an algorithm that has recently been introduced by Hazan. We present a refined analysis of Hazan's algorithm that results in improved running time bounds for a single solution as well as for the whole solution path as a function of the approximation guarantee.

ISBN:978-3-8439-0387-5