Fakultät für Mathematik, Informatik und Statistik - Digitale Hochschulschriften der LMU - Teil 02/02

Grasping objects with our hands allows us to skillfully move and manipulate them.

Hand-held tools further extend our capabilities by adapting precision, power, and shape of our hands to the task at hand.

Some of these tools, such as mobile phones or computer mice, already incorporate information processing capabilities.

Many other tools may be augmented with small, energy-efficient digital sensors and processors.

This allows for graspable objects to learn about the user grasping them - and supporting the user's goals.

For example, the way we grasp a mobile phone might indicate whether we want to take a photo or call a friend with it - and thus serve as a shortcut to that action.

A power drill might sense whether the user is grasping it firmly enough and refuse to turn on if this is not the case.

And a computer mouse could distinguish between intentional and unintentional movement and ignore the latter.

This dissertation gives an overview of grasp sensing for human-computer interaction, focusing on technologies for building grasp-sensitive surfaces and challenges in designing grasp-sensitive user interfaces.

It comprises three major contributions: a comprehensive review of existing research on human grasping and grasp sensing, a detailed description of three novel prototyping tools for grasp-sensitive surfaces, and a framework for analyzing and designing grasp interaction:

For nearly a century, scientists have analyzed human grasping.

My literature review gives an overview of definitions, classifications, and models of human grasping.

A small number of studies have investigated grasping in everyday situations.

They found a much greater diversity of grasps than described by existing taxonomies.

This diversity makes it difficult to directly associate certain grasps with users' goals.

In order to structure related work and own research, I formalize a generic workflow for grasp sensing.

It comprises *capturing* of sensor values, *identifying* the associated grasp, and *interpreting* the meaning of the grasp.

A comprehensive overview of related work shows that implementation of grasp-sensitive surfaces is still hard, researchers often are not aware of related work from other disciplines, and intuitive grasp interaction has not yet received much attention.

In order to address the first issue, I developed three novel sensor technologies designed for grasp-sensitive surfaces. These mitigate one or more limitations of traditional sensing techniques:

**HandSense** uses four strategically positioned capacitive sensors for detecting and classifying grasp patterns on mobile phones. The use of custom-built high-resolution sensors allows detecting proximity and avoids the need to cover the whole device surface with sensors. User tests showed a recognition rate of 81%, comparable to that of a system with 72 binary sensors.

**FlyEye** uses optical fiber bundles connected to a camera for detecting touch and proximity on arbitrarily shaped surfaces. It allows rapid prototyping of touch- and grasp-sensitive objects and requires only very limited electronics knowledge. For FlyEye I developed a *relative calibration* algorithm that allows determining the locations of groups of sensors whose arrangement is not known.

**TDRtouch** extends Time Domain Reflectometry (TDR), a technique traditionally used for inspecting cable faults, for touch and grasp sensing. TDRtouch is able to locate touches along a wire, allowing designers to rapidly prototype and implement modular, extremely thin, and flexible grasp-sensitive surfaces.

I summarize how these technologies cater to different requirements and significantly expand the design space for grasp-sensitive objects.

Furthermore, I discuss challenges for making sense of raw grasp information and categorize interactions. Traditional application scenarios for grasp sensing use only the grasp sensor's data, and only for mode-switching. I argue that data from gr

In recent years, we started to define our life through experiences we make instead of objectswe buy. To attend a concert of our favorite musician may be more important for us thanowning an expensive stereo system. Similarly, we define interactive systems not only by thequality of the display or its usability, but rather by the experiences we can make when usingthe device. A cell phone is primarily built for making calls and receiving text messages,but on an emotional level it might provide a way to be close to our loved ones, even thoughthey are far away sometimes. When designing interactive technology, we do not only haveto answer the question how people use our systems, but also why they use them. Thus,we need to concentrate on experiences, feelings and emotions arising during interaction.Experience Design is an approach focusing on the story that a product communicates beforeimplementing the system.

In an interdisciplinary team of psychologists, industrial designers, product developers andspecialists in human-computer interaction, we applied an Experience Design process to theautomotive domain. A major challenge for car manufacturers is the preservation of theseexperiences throughout the development process. When implementing interactive systemsengineers rely on technical requirements and a set of constraints (e.g., safety) oftentimescontradicting aspects of the designed experience. To resolve this conflict, Experience Prototypingis an important tool translating experience stories to an actual interactive product.

With this thesis I investigate the Experience Design process focusing on Experience Prototyping.Within the automotive context, I report on three case studies implementing threekinds of interactive systems, forming and following our approach. I implemented (1) anelectric vehicle information system called Heartbeat, communicating the state of the electricdrive and the batteries to the driver in an unobtrusive and ensuring way. I integrated Heartbeatinto the dashboard of a car mock-up with respect to safety and space requirements butat the same time holding on to the story in order to achieve a consistent experience. With (2)the Periscope I implemented a mobile navigation device enhancing the social and relatednessexperiences of the passengers in the car. I built and evaluated several experience prototypesin different stages of the design process and showed that they transported the designed experiencethroughout the implementation of the system. Focusing on (3) the experience offreehand gestures, GestShare explored this interaction style for in-car and car-to-car socialexperiences. We designed and implemented a gestural prototypes for small but effectivesocial interactions between drivers and evaluated the system in the lab and and in-situ study.

The contributions of this thesis are (1) a definition of Experience Prototyping in the automotivedomain resulting from a literature review and my own work, showing the importanceand feasibility of Experience Prototyping for Experience Design. I (2) contribute three casestudies and describe the details of several prototypes as milestones on the way from a anexperience story to an interactive system. I (3) derive best practices for Experience Prototypingconcerning their characteristics such as fidelity, resolution and interactivity as well asthe evaluation in the lab an in situ in different stages of the process.

In today’s business world, decisions have to be made on different levels, including strategic, tactical, and operational levels. Decisions on the strategic level are characterized by their complexity, longevity and impact. Such decisions can benefit from the participation of a large, diverse group of people as they contribute different background knowledge, perspectives, and evaluation criteria. Typically, such decisions need to be considered over a prolonged period of time as opinions may need to be formed or may change due to the availability of new information. The goal of people in group decision making situations is typically to achieve good decisions. A mechanism is thus desirable that is capable of addressing the aforementioned challenges and of producing a good decision. For this work, a decision is thought to be good if it is predominantly based on the sincere opinions of the participants.

In this thesis, we investigate the market metaphor as a promising approach for group decision making. Markets are attributed with the capability of gathering and aggregating assessments from people in a single indicator, the price. They allow for a continued participation over a prolonged time, reversibility of one’s market position by repeated trading, and the usage of individual evaluation criteria. For investigating the application of the market metaphor to decision making, we develop LDM, a market-based approach for group decision making. There, we represent a pending decision as a market and the decision options as stocks. Participants then buy shares of their favored stocks and sell shares of the stocks they dislike. High demand leads to price increase whereas low prices are the result of low demand. The most favored decision options can be identified from the ranking of the stocks according to their prices. To support the achievement of a good decision, we model the market behavior of participants, devise design principles, identify suitable application scenarios, and determine appropriate functionalities for a market software. We furthermore devise the concept of market perturbations for uncovering the trading intentions of participants.

We furthermore implement a web-based software prototype of LDM. It provides functionalities for decision making, market trading, user handling, information exchange, and market perturbations. Participants there trade their favored stocks using virtual play money. We test the LDM approach and its software prototype in an EU-funded project, in a lab study, in the selection of research proposals, and in a university seminar for scenario building.

Expectile regression can be seen as an extension of available (mean) regression models as it describes more general properties of the response distribution. This thesis introduces to expectile regression and presents new extensions of existing semiparametric regression models.

The dissertation consists of four central parts. First, the one-to-one-connection between expectiles, the cumulative distribution function (cdf) and quantiles is used to calculate the cdf and quantiles from a fine grid of expectiles. Quantiles-from-expectiles-estimates are introduced and compared with direct quantile estimates regarding e�ciency. Second, a method to estimate non-crossing expectile curves based on splines is developed. Also, the case of clustered or longitudinal observations is handled by introducing random individual components which leads to an extension of mixed models to mixed expectile models. Third, quantiles-from-expectiles-estimates in the framework of unequal probability sampling are proposed.

All methods are implemented and available within the package expectreg via the open source software R. As fourth part, a description of the package expectreg is given at the end of this thesis.

Penalisierte Regressionsmodelle stellen eine Möglichkeit dar die Selektion von Kovariablen in die Schätzung eines Modells zu integrieren. Penalisierte Ansätze eignen sich insbesondere dafür, komplexen Strukturen in den Kovariablen eines Modells zu berücksichtigen. Diese Arbeit beschäftigt sich mit verschiedenen Penalisierungsansätzen für diskrete Strukturen, wobei der Begriff "diskrete Struktur" in dieser Arbeit alle Arten von kategorialen Einflussgrößen, von effekt-modifizierenden, kategorialen Einflussgrößen sowie von gruppenspezifischen Effekten in hierarchisch strukturierten Daten bezeichnet. Ihnen ist gemein, dass sie zu einer verhältnismäßig großen Anzahl an zu schätzenden Koeffizienten führen können. Deswegen besteht ein besonderes Interesse daran zu erfahren, welche Kategorien einer Einflussgröße die Zielgröße beeinflussen, und welche Kategorien unterschiedliche beziehungsweise ähnliche Effekte auf die Zielgröße haben. Kategorien mit ähnlichen Effekten können beispielsweise durch fused Lasso Penalties identifiziert werden. Jedoch beschränken sich einige, bestehende Ansätze auf das lineare Modell. Die vorliegende Arbeit überträgt diese Ansätze auf die Klasse der generalisierten linearen Regressionsmodelle. Das beinhaltet computationale wie theoretische Aspekte. Konkret wird eine fused Lasso Penalty für effekt-modifizierende kategoriale Einflussgrößen in generalisierten linearen Regressionsmodellen vorgeschlagen. Sie ermöglicht es, Einflussgrößen zu selektieren und Kategorien einer Einflussgröße zu fusionieren. Gruppenspezifische Effekte, die die Heterogenität in hierarchisch strukturierten Daten berücksichtigen, sind ein Spezialfall einer solchen effekt-modifizierenden, kategorialen Größe. Hier bietet der penalisierte Ansatz zwei wesentliche Vorteile: (i) Im Gegensatz zu gemischten Modellen, die stärkere Annahmen treffen, kann der Grad der Heterogenität sehr leicht reduziert werden. (ii) Die Schätzung ist effizienter als im unpenalisierten Ansatz. In orthonormalen Settings können Fused Lasso Penalties konzeptionelle Nachteile haben. Als Alternative wird eine L0 Penalty für diskrete Strukturen in generalisierten linearen Regressionsmodellen diskutiert, wobei die sogenannte L0 "Norm" eine Indikatorfunktion für Argumente ungleich Null bezeichnet. Als Penalty ist diese Funktion so interessant wie anspruchsvoll. Betrachtet man eine Approximation der L0 Norm als Verlustfunktion wird im Grenzwert der bedingte Modus einer Zielgröße geschätzt.

Financial bubbles have been present in the history of financial markets from the early days up to the modern age. An asset is said to exhibit a bubble when its market value exceeds its fundamental valuation. Although this phenomenon has been thoroughly studied in the economic literature, a mathematical martingale theory of bubbles, based on an absence of arbitrage has only recently been developed. In this dissertation, we aim to further contribute to the developement of this theory.

In the first part we construct a model that allows us to capture the birth of a financial bubble and to describe its behavior as an initial submartingale in the build-up phase, which then turns into a supermartingale in the collapse phase. To this purpose we construct a flow in the space of equivalent martingale measures and we study the shifting perception of the fundamental value of a given asset.

In the second part of the dissertation, we study the formation of financial bubbles in the valuation of defaultable claims in a reduced-form setting. In our model a bubble is born due to investor heterogeneity. Furthermore, our study shows how changes in the dynamics of the defaultable claim's market price may lead to a different selection of the martingale measure used for pricing. In this way we are able to unify the classical martingale theory of bubbles with a constructive approach to the study of bubbles, based on the interactions between investors.

An efficient way to represent the domain knowledge is relational data, where information is recorded in form of relationships between entities. Relational data is becoming ubiquitous over the years for knowledge representation due to the fact that many real-word data is inherently interlinked. Some well-known examples of relational data are: the World Wide Web (WWW), a system of interlinked hypertext documents; the Linked Open Data (LOD) cloud of the Semantic Web, a collection of published data and their interlinks; and finally the Internet of Things (IoT), a network of physical objects with internal states and communications ability. Relational data has been addressed by many different machine learning approaches, the most promising ones are in the area of relational learning, which is the focus of this thesis. While conventional machine learning algorithms consider entities as being independent instances randomly sampled from some statistical distribution and being represented as data points in a vector space, relational learning takes into account the overall network environment when predicting the label of an entity, an attribute value of an entity or the existence of a relationship between entities. An important feature is that relational learning can exploit contextual information that is more distant in the relational network. As the volume and structural complexity of the relational data increase constantly in the era of Big Data, scalability and the modeling power become crucial for relational learning algorithms.

Previous relational learning algorithms either provide an intuitive representation of the model, such as Inductive Logic Programming (ILP) and Markov Logic Networks (MLNs), or assume a set of latent variables to explain the observed data, such as the Infinite Hidden Relational Model (IHRM), the Infinite Relational Model (IRM) and factorization approaches. Models with intuitive representations often involve some form of structure learning which leads to scalability problems due to a typically large search space. Factorizations are among the best-performing approaches for large-scale relational learning since the algebraic computations can easily be parallelized and since they can exploit data sparsity. Previous factorization approaches exploit only patterns in the relational data itself and the focus of the thesis is to investigate how additional prior information (comprehensive information), either in form of unstructured data (e.g., texts) or structured patterns (e.g., in form of rules) can be considered in the factorization approaches. The goal is to enhance the predictive power of factorization approaches by involving prior knowledge for the learning, and on the other hand to reduce the model complexity for efficient learning.

This thesis contains two main contributions:

The first contribution presents a general and novel framework for predicting relationships in multirelational data using a set of matrices describing the various instantiated relations in the network. The instantiated relations, derived or learnt from prior knowledge, are integrated as entities' attributes or entity-pairs' attributes into different adjacency matrices for the learning. All the information available is then combined in an additive way. Efficient learning is achieved using an alternating least squares approach exploiting sparse matrix algebra and low-rank approximation. As an illustration, several algorithms are proposed to include information extraction, deductive reasoning and contextual information in matrix factorizations for the Semantic Web scenario and for recommendation systems. Experiments on various data sets are conducted for each proposed algorithm to show the improvement in predictive power by combining matrix factorizations with prior knowledge in a modular way.

In contrast to a matrix, a 3-way tensor si a more natural representation for the multirelational data where entities are connected by diffe

Forest-fire processes were first introduced in the physics literature as a toy model for self-organized criticality. The term self-organized criticality describes interacting particle systems which are governed by local interactions and are inherently driven towards a perpetual critical state. As in equilibrium statistical physics, the critical state is characterized by long-range correlations, power laws, fractal structures and self-similarity.

We study several different forest-fire models, whose common features are the following: All models are continuous-time processes on the vertices of some graph. Every vertex can be "vacant" or "occupied by a tree". We start with some initial configuration. Then the process is governed by two competing random mechanisms: On the one hand, vertices become occupied according to rate 1 Poisson processes, independently of one another. On the other hand, occupied clusters are "set on fire" according to some predefined rule. In this case the entire cluster is instantaneously destroyed, i.e. all of its vertices become vacant.

The self-organized critical behaviour of forest-fire models can only occur on infinite graphs such as planar lattices or infinite trees. However, in all relevant versions of forest-fire models, the destruction mechanism is a priori only well-defined for finite graphs. For this reason, one starts with a forest-fire model on finite subsets of an infinite graph and then takes the limit along increasing sequences of finite subsets to obtain a new forest-fire model on the infinite graph.

In this thesis, we perform this kind of limit for two classes of forest-fire models and investigate the resulting limit processes.