Fakultät für Mathematik, Informatik und Statistik - Digitale Hochschulschriften der LMU - Teil 01/02 Archiv

The development of highly interactive multimedia applications is still a challenging and complex task. In addition to the application logic, multimedia applications typically provide a sophisticated user interface with integrated media objects. As a consequence, the development process involves different experts for software design, user interface design, and media design. There is still a lack of concepts for a systematic development which integrates these aspects. This thesis provides a model-driven development approach addressing this problem. Therefore it introduces the Multimedia Modeling Language (MML), a visual modeling language supporting a design phase in multimedia application development. The language is oriented on well-established software engineering concepts, like UML 2, and integrates concepts from the areas of multimedia development and model-based user interface development. MML allows the generation of code skeletons from the models. Thereby, the core idea is to generate code skeletons which can be directly processed in multimedia authoring tools. In this way, the strengths of both are combined: Authoring tools are used to perform the creative development tasks while models are used to design the overall application structure and to enable a well-coordinated development process. This is demonstrated using the professional authoring tool Adobe Flash. MML is supported by modeling and code generation tools which have been used to validate the approach over several years in various student projects and teaching courses. Additional prototypes have been developed to demonstrate, e.g., the ability to generate code for different target platforms. Finally, it is discussed how models can contribute in general to a better integration of well-structured software development and creative visual design.

The increasing importance of distributed and decentralized software architectures entails more and more attention for adaptive software. Obtaining adaptiveness, however, is a difficult task as the software design needs to foresee and cope with a variety of situations. Using reconfiguration of components facilitates this task, as the adaptivity is conducted on an architecture level instead of directly in the code. This results in a separation of concerns; the appropriate reconfiguration can be devised on a coarse level, while the implementation of the components can remain largely unaware of reconfiguration scenarios. We study reconfiguration in component frameworks based on formal theory. We first discuss programming with components, exemplified with the development of the cmc model checker. This highly efficient model checker is made of C++ components and serves as an example for component-based software development practice in general, and also provides insights into the principles of adaptivity. However, the component model focuses on high performance and is not geared towards using the structuring principle of components for controlled reconfiguration. We thus complement this highly optimized model by a message passing-based component model which takes reconfigurability to be its central principle. Supporting reconfiguration in a framework is about alleviating the programmer from caring about the peculiarities as much as possible. We utilize the formal description of the component model to provide an algorithm for reconfiguration that retains as much flexibility as possible, while avoiding most problems that arise due to concurrency. This algorithm is embedded in a general four-stage adaptivity model inspired by physical control loops. The reconfiguration is devised to work with stateful components, retaining their data and unprocessed messages. Reconfiguration plans, which are provided with a formal semantics, form the input of the reconfiguration algorithm. We show that the algorithm achieves perceived atomicity of the reconfiguration process for an important class of plans, i.e., the whole process of reconfiguration is perceived as one atomic step, while minimizing the use of blocking of components. We illustrate the applicability of our approach to reconfiguration by providing several examples like fault-tolerance and automated resource control.

This dissertation deals with type systems which guarantee polynomial time complexity of typed programs. Such algorithms are commonly regarded as being feasible for practical applications, because their runtime grows reasonably fast for bigger inputs. The implicit complexity community has proposed several type systems for polynomial time in the recent years, each with strong, but different structural restrictions on the permissible algorithms which are necessary to control complexity. Comparisons between the various approaches are hard and this has led to a landscape of islands in the literature of expressible algorithms in each calculus, without many known links between them. This work chooses Light Affine Logic (LAL) and Hofmann's LFPL, both linearly typed, and studies the connections between them. It is shown that the light iteration in LAL, the fixed point variant of LAL, is expressive enough to allow a (non-trivial) compositional embedding of LFPL. The pull-out trick of LAL is identified as a technique to type certain non-size-increasing algorithms in such a way that they can be iterated. The System T sibling of LAL is developed which seamlessly integrates this technique as a central feature of the iteration scheme and which is proved again correct and complete for polynomial time. Because -iterations of the same level cannot be nested, is further generalised to , which surprisingly can express the impredicative iteration of LFPL and the light iteration of at the same time. Therefore, it subsumes both systems in one, while still being polynomial time normalisable. Hence, this result gives the first bridge between these two islands of implicit computational complexity.

Decision theory is, in particular in economics, medical expert systems and statistics, an important tool for determining optimal decisions under uncertainty. In view of applications in statistics, the present book is concerned with decision problems which are explicitly data-based. Since the arising uncertainties are often too complex to be described by classical precise probability assessments, concepts of imprecise probabilities (coherent lower previsions, F-probabilities) are applied. Due to the present state of research, some basic groundwork has to be done: Firstly, topological properties of different concepts of imprecise probabilities are investigated. In particular, the concept of coherent lower previsions appears to have advantageous properties for applications in decision theory. Secondly, several decision theoretic tools are developed for imprecise probabilities. These tools are mainly based on concepts developed by L. Le Cam and enable, for example, a definition of sufficiency in case of imprecise probabilities for the first time. Building on that, the article [A. Buja, Zeitschrift für Wahrscheinlichkeitstheorie und Verwandte Gebiete 65 (1984) 367-384] is reinvestigated in the only recently available framework of imprecise probabilities. This leads to a generalization of results within the Huber-Strassen theory concerning least favorable pairs or models. Results obtained by these investigations can also be applied afterwards in order to justify the use of the method of natural extension, which is fundamental within the theory of imprecise probabilities, in data-based decision problems. It is shown by means of the theory of vector lattices that applying the method of natural extension in decision problems does not affect the optimality of decisions. However, it is also shown that, in general, the method of natural extension suffers from a severe instability. The book closes with an application in statistics in which a minimum distance estimator is developed for imprecise probabilities. After an investigation concerning its asymptotic properties, an algorithm for calculating the estimator is given which is based on linear programming. This algorithm has led to an implementation of the estimator in the programming language R which is publicly available as R package "imprProbEst". The applicability of the estimator (even for large sample sizes) is demonstrated in a simulation study.

The development of high-throughput techniques has lead to a paradigm change in biology from the small-scale analysis of individual genes and proteins to a genome-scale analysis of biological systems. Proteins and genes can now be studied in their interaction with each other and the cooperation within multi-subunit protein complexes can be investigated. Moreover, time-dependent dynamics and regulation of these processes and associations can now be explored by monitoring mRNA changes and turnover. The in-depth analysis of these large and complex data sets would not be possible without sophisticated algorithms for integrating different data sources, identifying interesting patterns in the data and addressing the high variability and error rates in biological measurements. In this thesis, we developed such methods for the investigation of protein interactions and complexes and the corresponding regulatory processes. In the first part, we analyze networks of physical protein-protein interactions measured in large-scale experiments. We show that the topology of the complete interactomes can be confidently extrapolated despite high numbers of missing and wrong interactions from only partial measurements of interaction networks. Furthermore, we find that the structure and stability of protein interaction networks is not only influenced by the degree distribution of the network but also considerably by the suppression or propagation of interactions between highly connected proteins. As analysis of network topology is generally focused on large eukaryotic networks, we developed new methods to analyze smaller networks of intraviral and virus-host interactions. By comparing interactomes of related herpesviral species, we could detect a conserved core of protein interactions and could address the low coverage of the yeast two-hybrid system. In addition, common strategies in the interaction of the viruses with the host cell were identified. New affinity purification methods now make it possible to directly study associations of proteins in complexes. Due to experimental errors the individual protein complexes have to be predicted with computational methods from these purification results. As previously published methods relied more or less heavily on existing knowledge on complexes, we developed an unsupervised prediction algorithm which is independent from such additional data. Using this approach, high-quality protein complexes can be identified from the raw purification data alone for any species purification experiments are performed. To identify the direct, physical interactions within these predicted complexes and their subcomponent structure, we describe a new approach to extract the highest scoring subnetwork connecting the complex and interactions not explained by alternative paths of indirect interactions. In this way, important interactions within the complexes can be identified and their substructure can be resolved in a straightforward way. To explore the regulation of proteins and complexes, we analyzed microarray measurements of mRNA abundance, de novo transcription and decay. Based on the relationship between newly transcribed, pre-existing and total RNA, transcript half-life can be estimated for individual genes using a new microarray normalization method and a quality control can be applied. We show that precise measurements of RNA half-life can be obtained from de novo transcription which are of superior accuracy to previously published results from RNA decay. Using such precise measurements, we studied RNA half-lives in human B-cells and mouse fibroblasts to identify conserved patterns governing RNA turnover. Our results show that transcript half-lives are strongly conserved and specifically correlated to gene function. Although transcript half-life is highly similar in protein complexes and \mbox{families}, individual proteins may deviate significantly from the remaining complex subunits or family members to efficiently support the regulation of protein complexes or to create non-redundant roles of functionally similar proteins. These results illustrate several of the many ways in which high-throughput measurements lead to a better understanding of biological systems. By studying large-scale measure\-ments in this thesis, the structure of protein interaction networks and protein complexes could be better characterized, important interactions and conserved strategies for herpes\-viral infection could be identified and interesting insights could be gained into the regulation of important biological processes and protein complexes. This was made possible by the development of novel algorithms and analysis approaches which will also be valuable for further research on these topics.

In den letzten Jahren gab es in verschiedensten Bereichen der Biologie einen dramatischen Anstieg verfügbarer, experimenteller Daten. Diese erlauben zum ersten Mal eine detailierte Analyse der Funktionsweisen von zellulären Komponenten wie Genen und Proteinen, die Analyse ihrer Verknüpfung in zellulären Netzwerken sowie der Geschichte ihrer Evolution. Insbesondere der Bioinformatik kommt hier eine wichtige Rolle in der Datenaufbereitung und ihrer biologischen Interpretation zu. In der vorliegenden Doktorarbeit werden zwei wichtige Bereiche der aktuellen bioinformatischen Forschung untersucht, nämlich die Analyse von Proteinstrukturevolution und Ähnlichkeiten zwischen Proteinstrukturen, sowie die Analyse von alternativem Splicing, einem integralen Prozess in eukaryotischen Zellen, der zur funktionellen Diversität beiträgt. Insbesondere führen wir mit dieser Arbeit die Idee einer kombinierten Analyse der beiden Mechanismen (Strukturevolution und Splicing) ein. Wir zeigen, dass sich durch eine kombinierte Betrachtung neue Einsichten gewinnen lassen, wie Strukturevolution und alternatives Splicing sowie eine Kopplung beider Mechanismen zu funktioneller und struktureller Komplexität in höheren Organismen beitragen. Die in der Arbeit vorgestellten Methoden, Hypothesen und Ergebnisse können dabei einen Beitrag zu unserem Verständnis der Funktionsweise von Strukturevolution und alternativem Splicing bei der Entstehung komplexer Organismen leisten wodurch beide, traditionell getrennte Bereiche der Bioinformatik in Zukunft voneinander profitieren können.

In this thesis we consider recent developments in insurance and electricity financial products. In particular, we investigate the interplay between insurance and finance, and therein the problem of pricing catastrophe insurance options written on a loss index as well as electricity products. Catastrophe insurance options are standardized exchange-traded financial securities based on an underlying index, e.g. a PCS index, that encompasses insurance losses due to natural catastrophes. The PCS index is provided by the Property Claim Services (PCS), a US independent industry authority which estimates catastrophic property damage. The advantages of the catastrophe options in comparison to other capital market insurance solutions are lower transaction costs relative to the re-insurance and minimal credit risk, because of the guarantee of the exchange. The main results of the thesis are fairly realistic models for catastrophe loss indexes and electricity futures markets, where by employing Fourier transform techniques we are able to provide analytical pricing formulas for European type options traded in the markets. For the catastrophe loss index we specify a model, where the initial estimate of each catastrophe loss is re-estimated immediately by a positive martingale starting from the random time of loss occurrence. Significant advantage of this methodology is that it can be applied to loss distributions with heavy tails -- the appropriate tail behavior for catastrophe modeling. The case when the re-estimation factors are given by positive affine martingales is also discussed and a characterization of positive affine local martingales is provided. For electricity futures markets we derive a model, where we can simultaneously model evolution of futures and spot prices. At the same time we have an explicit connection between electricity futures and spot price processes. Furthermore, an important achievement is that the spot price dynamics in this model becomes multi-dimensional Markovian. The Markovian structure is crucial for pricing of path dependent electricity options.

This thesis describes novel processes in two important areas of human-computer interaction (HCI) and demonstrates ways to combine these in appropriate ways. First, prototyping plays an essential role in the development of complex applications. This is especially true if a user-centred design process is followed. We describe and compare a set of existing toolkits and frameworks that support the development of prototypes in the area of pervasive computing. Based on these observations, we introduce the EIToolkit that allows the quick generation of mobile and pervasive applications, and approaches many issues found in previous works. Its application and use is demonstrated in several projects that base on the architecture and an implementation of the toolkit. Second, we present novel results and extensions in user modelling, specifically for predicting time to completion of tasks. We extended established concepts such as the Keystroke-Level Model to novel types of interaction with mobile devices, e.g. using optical markers and gestures. The design, creation, as well as a validation of this model are presented in some detail in order to show its use and usefulness for making usability predictions. The third part is concerned with the combination of both concepts, i.e. how to integrate user models into the design process of pervasive applications. We first examine current ways of developing and show generic approaches to this problem. This leads to a concrete implementation of such a solution. An innovative integrated development environment is provided that allows for quickly developing mobile applications, supports the automatic generation of user models, and helps in applying these models early in the design process. This can considerably ease the process of model creation and can replace some types of costly user studies.

The emergence of event-driven architectures, automation of business processes, drastic cost-reductions in sensor technology, and a growing need to monitor IT systems (as well as other systems) due to legal, contractual, or operational considerations lead to an increasing generation of events. This development is accompanied by a growing demand for managing and processing events in an automated and systematic way. Complex Event Processing (CEP) encompasses the (automatable) tasks involved in making sense of all events in a system by deriving higher-level knowledge from lower-level events while the events occur, i.e., in a timely, online fashion and permanently. At the core of CEP are queries which monitor streams of "simple" events for so-called complex events, that is, events or situations that manifest themselves in certain combinations of several events occurring (or not occurring) over time and that cannot be detected from looking only at single events. Querying events is fundamentally different from traditional querying and reasoning with database or Web data, since event queries are standing queries that are evaluated permanently over time against incoming streams of event data. In order to express complex events that are of interest to a particular application or user in a convenient, concise, cost-effective and maintainable manner, special purpose Event Query Languages (EQLs) are needed. This thesis investigates practical and theoretical issues related to querying complex events, covering the spectrum from language design over declarative semantics to operational semantics for incremental query evaluation. Its central topic is the development of the high-level event query language XChangeEQ. In contrast to previous data stream and event query languages, XChangeEQ's language design recognizes the four querying dimensions of data extractions, event composition, temporal relationships, and, for non-monotonic queries involving negation or aggregation, event accumulation. XChangeEQ deals with complex structured data in event messages, thus addressing the need to query events communicated in XML formats over the Web. It supports deductive rules as an abstraction and reasoning mechanism for events. To achieve a full coverage of the four querying dimensions, it builds upon a separation of concerns of the four querying dimensions, which makes it easy-to-use and highly expressive. A recurrent theme in the formal foundations of XChangeEQ is that, despite the fundamental differences between traditional database queries and event queries, many well-known results from databases and logic programming are, with some importance changes, applicable to event queries. Declarative semantics for XChangeEQ are given as a (Tarski-style) model theory with accompanying fixpoint theory. This approach accounts well for (1) data in events and (2) deductive rules defining new events from existing ones, two aspects often neglected in previous work of semantics of EQLs. For the evaluation of event queries, this work introduces operational semantics based on an extended and tailored form of relational algebra and query plans with materialization points. Materialization points account for storing and maintaining information about those received events that are relevant for, i.e., can contribute to, future query answers, as well as for an incremental evaluation that avoids recomputing certain intermediate results. Efficient state maintenance in incremental evaluation is approached by "differentiating" algebra expressions, i.e., by deriving expressions for computing only the changes to materialization points. Knowing how long an event is relevant is a prerequisite for performing garbage collection during event query evaluation and also of central importance for developing cost-based query planners. To this end, this thesis introduces a notion of relevance of events (to a given query plan) and develops methods for determining temporal relevance, a particularly useful form based on time-related information.