05 Fakultät Informatik, Elektrotechnik und Informationstechnik
Permanent URI for this collectionhttps://elib.uni-stuttgart.de/handle/11682/6
Browse
5 results
Search Results
Item Open Access Data-efficient and safe learning with Gaussian processes(2020) Schreiter, Jens; Toussaint, Marc (Prof. Dr. rer. nat.)Data-based modeling techniques enjoy increasing popularity in many areas of science and technology where traditional approaches are limited regarding accuracy and efficiency. When employing machine learning methods to generate models of dynamic system, it is necessary to consider two important issues. Firstly, the data-sampling process should induce an informative and representative set of points to enable high generalization accuracy of the learned models. Secondly, the algorithmic part for efficient model building is essential for applicability, usability, and the quality of the learned predictive model. This thesis deals with both of these aspects for supervised learning problems, where the interaction between them is exploited to realize an exact and powerful modeling. After introducing the non-parametric Bayesian modeling approach with Gaussian processes and basics for transient modeling tasks in the next chapter, we dedicate ourselves to extensions of this probabilistic technique to relevant practical requirements in the subsequent chapter. This chapter provides an overview on existing sparse Gaussian process approximations and propose some novel work to increase efficiency and model selection on particularly large training data sets. For example, our sparse modeling approach enables real-time capable prediction performance and efficient learning with low memory requirements. A comprehensive comparison on various real-world problems confirms the proposed contributions and shows a variety of modeling tasks, where approximate Gaussian processes can be successfully applied. Further experiments provide more insight about the whole learning process, and thus a profound understanding of the presented work. In the fourth chapter, we focus on active learning schemes for safe and information-optimal generation of meaningful data sets. In addition to the exploration behavior of the active learner, the safety issue is considered in our work, since interacting with real systems should not result in damages or even completely destroy it. Here we propose a new model-based active learning framework to solve both tasks simultaneously. As basis for the data-sampling process we employ the presented Gaussian process techniques. Furthermore, we distinguish between static and transient experimental design strategies. Both problems are separately considered in this chapter. Nevertheless, the requirements for each active learning problem are the same. This subdivision into a static and transient setting allows a more problem-specific perspective on the two cases, and thus enables the creation of specially adapted active learning algorithms. Our novel approaches are then investigated for different applications, where a favorable trade-off between safety and exploration is always realized. Theoretical results maintain these evaluations and provide respectable knowledge about the derived model-based active learning schemes. For example, an upper bound for the probability of failure of the presented active learning methods is derived under reasonable assumptions. Finally, the thesis concludes with a summary of the investigated machine learning problems and motivate some future research directions.Item Open Access On consistency and distribution in software-defined networking(2020) Kohler, Thomas; Rothermel, Kurt (Prof. Dr. rer. nat. Dr. h. c.)Software-defined Networking (SDN) is an emerging networking paradigm promising flexible programmability and simplified management. Over the last years, SDN has built up huge momentum in academia that has led to huge practical impact through the large-scale adoption of big industrial players like Google, Facebook, and Microsoft driving cloud computing, data center networks, and their interconnection in SDN-based wide-area networks. SDN is a key enabler for high dynamics in terms of network reconfiguration and innovation, allowing the deployment of new network protocols and substantially expanding the networking paradigm by moving applications into the network, both at unprecedented pace and ease. The SDN paradigm is centered around the separation of the data plane from the logically centralized but typically physically distributed control plane that programs the forwarding behaviour of the network devices in the data plane based on a global view. Especially requirements on correctness, scalability, availability, and resiliency raised through practical adoption at scale have put a strong emphasis on consistency and distribution in the SDN paradigm. This thesis addresses various challenges regarding consistency and distribution in Software-defined Networking. More specifically, it focusses and contributes to the research areas of update consistency, flexibility in control plane distribution, and data plane implementation of a distributed application. Reconfiguring an SDN-based network inevitably requires to update the rules that determine the forwarding behaviour of the devices in its data plane. Updating these rules, which are situated on the inherently distributed data plane devices, is an asynchronous process. Hence, packets traversing the network may be processed according to a mixture of new and old rules during the update process. Consequently arising inconsistency effects can severely degrade the network performance and can break stipulated network invariants for instance on connectivity or security. We introduce a general architecture for network management under awareness of expectable update-induced inconsistency effects, which allows for an appropriate selection of an update mechanism and its parameters in order to prevent those effects. We thoroughly analyze update consistency for the case of multicast networks, show crucial particularities and present mechanisms for the prevention and mitigation of multicast-specific inconsistency effects. Observing that on the one hand SDN's separation of control has been deemed rather strict, moving any control ``intelligence'' from the data plane devices to remote controller entities hence increasing control latency while on the other hand the coupling between controller and data plane devices is quite tight hence hindering free distribution of control logic, we present a controller architecture enabling flexible and full-range distribution of network control. The architecture is based on decoupling through an event abstraction and a flexible dissemination scheme for those events based on the content-based publish/subscribe paradigm. This lightweight design allows to push down control logic back onto data plane devices. Thus, we expand SDN's control paradigm and enable the full range from fully decentralized control, over local control still profiting from global view up to fully centralized control. This scheme allows to trade-off scope of state data, consistency semantics and synchronization overhead, control latency, and quality of control decisions. Furthermore, our implementation covers a large set of mechanisms for improving control plane consistency and scalability, such as inherent load-balancing, fast autonomous control decision making, detection of policy conflicts, and a feedback mechanism for data plane updates. In a last area, we focus on the implementation of a distributed application from the domain of message-oriented middleware in the data plane. We implement Complex Event Processing (CEP) on top of programmable network devices employing data plane programming, a recent big trend in SDN, or more specifically, using the P4 language. We discuss challenges entailed in the distributed data plane processing and address aspects of distribution and consistency in particular regarding consistency in stateful data plane programming, where internal state that determines how packets are processed is changed within this very processing, in turn changing the processing of subsequent packets. Since packet processing is executed in parallel on different execution units on the same device sharing the same state data, strong consistency semantics are required in order to ensure application correctness. Enabled by P4's flexible and powerful programming model, our data plane implementation of CEP yields greatly reduced latency and increased throughput. It comprises a compiler that compiles patterns for the detection of complex events specified in our rule specification language to P4 programs, consisting of a state machine and operators that process so-called windows containing historic events.Item Open Access Utilizing networked mobile devices for scientific simulations(2020) Dibak, Christoph; Rothermel, Kurt (Prof. Dr.)Numerical simulations on mobile devices create new applications supporting engineers and scientists in the field. Boosted by novel augmented reality devices, in-field analysis of complex systems allow engineers to make better decisions and predict the behavior of such systems by assuming different parameters before making risky and costly decisions. Mobile simulations are challenging as battery-powered mobile devices are only equipped with slow processors and are limited in energy resources. At the same time, mobile devices are only connected via wireless communication subjected to environmental conditions that might cause slow bandwidths or even disconnections to remote computing resources. Nevertheless, concepts presented in this thesis assume a distributed computation between mobile device and a powerful remote server. This thesis covers three major areas of the research field of mobile simulations. First, it provides concepts for distributed execution between server and mobile device in case of frequent disconnections. Second, it provides concepts using computationally less complex surrogate models for faster computation on the mobile device while still utilizing remote resources. Third, it provides concepts utilizing model order reduction for fast execution on mobile devices by pre-computing and adaptation of reduced models on a connected server. Evaluations show that concepts presented in this thesis significantly increase the performance of mobile simulations. In the case of disconnections, the number of deadline misses is reduced by 61 % while reducing the energy consumption by more than 74 % compared to a simplified approach. Concepts utilizing surrogate models speed-up the computation of the simulation by a factor of 6.5. Lastly, concepts utilizing model order reduction reduce the time for the computation of simulation results by a factor of 131 while using 73 times less energy for the specific test application.Item Open Access A model-based approach for data processing in IoT environments(2020) Franco da Silva, Ana Cristina; Mitschang, Bernhard (Prof. Dr.-Ing. habil.)The recent advances in several areas, including sensor technologies, networking, and data processing, have enabled the Internet of Things (IoT) vision to become more and more a reality every day. As a consequence of these advances, the IoT of today allows the development of sophisticated applications for IoT environments, such as smart cities, smart homes, or smart factories. Due to continuous sensor measurements and frequent data exchange among so-called IoT objects, the data generated within an IoT environment incorporate the form of data streams. With this increasing amount of data to be continuously processed, several challenges arise while aiming at an efficient processing of IoT data. For instance, how IoT data processing can be realized, so that meaningful information can be derived without affecting the reactiveness of IoT applications. Furthermore, how different functional, non-functional, and user-defined requirements of IoT applications can be satisfied by the IoT data processing. In this PhD thesis, a new holistic approach for processing data stream-based applications within IoT environments is presented. Its focus lies on efficient placement of operators of data stream applications onto heterogeneous, distributed, dynamic IoT environments. In contrast to state-of-the-art operator placement, this approach takes into consideration additional requirements introduced by the peculiar characteristics of the Internet of Things. Furthermore, non-functional and user-defined requirements are also taken into consideration. This PhD thesis is supported by different informational models and operator placement techniques, so that the entire life cycle of IoT environments and data stream-based applications can be easily managed. IoT environments and their processing capabilities are described by IoT environment models (IoTEM). Likewise, the business logic of IoT applications and their requirements are defined by data stream processing models (DSPM). Based on these informational models, several algorithms determine feasible placements of processing operators onto IoT objects of IoT environments, so that the aforementioned requirements and capabilities are matched. In this approach, one of the main goals is to process IoT data as near to data sources as possible, so that cloud infrastructures are employed only in cases where IoT environments do not offer sufficient processing resources for the IoT application. The execution of data processing on both IoT environments and cloud infrastructures is commonly known as fog computing. Through the approach of this PhD thesis, data processing of IoT applications can be tailored to particular use cases, supporting the specific requirements of the domains, and furthermore, of IoT application users. Once feasible placements are determined, processing operators are then deployed onto corresponding IoT objects using standards, such as TOSCA, and the IoT application is considered up and running. Finally, the IoT environment is continuously monitored in order to recognize and react to disturbances affecting the data processing of deployed IoT applications. The approach of this PhD thesis is supported by the Multi-purpose Binding and Provisioning Platform (MBP), an open-source IoT platform, which has been developed as a proof-of-concept of the contributions of this PhD thesis.Item Open Access Quality assessment in text analysis pipelines(2020) Kiefer, Cornelia; Mitschang, Bernhard (Prof. Dr.-Ing. habil.)High quality data and data analysis results are a precondition for future concepts such as the data-driven factory of the future. The quality of business decisions is directly influenced by the quality of data and analysis results. Current data quality concepts and tools only consider the raw input data of data analysis pipelines. They fail to regard specifics of analysis tools as well as data for each step of analysis pipelines. To fill this research gap, the QUALM concept for continuous and holistic data quality measurement and improvement within data analysis pipelines is presented in this thesis. In QUALM, data characteristics as well as specifics of analysis tools such as training data, features and semantic resources are regarded in each step of analysis pipelines. Existing data quality metrics measure the data quality of structured data, e.g., by counting null values, duplicates or invalid values. Equivalent approaches for textual data are missing. Additionally, most domain-specific text data sets are unlabeled. Thus, in addition to missing data quality metrics, also evaluation metrics are not calculable for these data sets and the thereupon derived analysis results. This leads to a high uncertainty of the analysts with respect to the quality of data and analysis results. QUALM conquers this challenge with a set of concrete text data quality methods. QUALM data quality indicators quantify text characteristics and give hints with respect to the expected quality of analysis results. Just as existing metrics for structured data determine, e.g., the number of null values and invalid fields, the QUALM indicators characterize texts with respect to, e.g., the number of abbreviations, spelling mistakes and ungrammatical sentences. Moreover, as demanded by the QUALM concept, these methods do not only consider the raw data, but also respect the specifics of the analysis tools. For example, QUALM has indicators which measure the confidence of standard analysis tools or the fit of semantic resources employed by analysis tools. Each indicator comes with a corresponding modifier. For example, the amount of abbreviations or spelling mistakes may be measured by a QUALM indicator. A corresponding QUALM modifier, e.g., modifies the data by means of resolving abbreviations or by a correction of spelling mistakes. Moreover, the selection of appropriate training data is especially difficult for analysts such as domain experts with little IT and/or data science knowledge. Yet, the appropriate selection of training data has a high impact on the quality of analysis results. Therefore, QUALM addresses this issue through a concrete method. The corresponding QUALM indicator measures data quality by means of the similarity between input and training data. In the case of textual data, text similarity metrics such as Latent Semantic Analysis and Cosine Similarity are employed. The counterpart QUALM modifier automatically selects the best-fitting training data and thus impedes low-quality results of domain-specific analysis. Finally, QUALM has another method which addresses data quality issues that arise from information extraction approaches that only consider either structured or unstructured text data in isolation. These isolated approaches may lead to a loss in terms of the amount of new information that may be presented to the analyst. In this thesis, this issue is addressed by a hybrid approach, which exploits structured and unstructured information sources in information extraction. To this end, especially structured data is considered which is enriched by unstructured free text fields. In the suggested approach, structured data is used to guide and improve the text analysis process. To this end, structured data is employed as a basis for a first grouping of free text fields and for removing information from the free texts which is already present in the structured fields. Thus, the hybrid approach yields more new and relevant information. The QUALM concept and methods are evaluated with respect to several industry-near application scenarios and corresponding concrete data sets. For example, the analysis of downtimes on a production line is considered. To this end a confidential industry data set comprising structured data enriched with free-text fields is employed. In further application scenarios, sample citizen data scientists are considered, i.e., domain experts with little IT and data science knowledge, who want to build analysis pipelines from scratch. E.g., they want to know customer opinions on a product. The evaluation results are very promising. The QUALM indicators and analysis result quality, measured as accuracy, correlate. Thus QUALM indicators are valid means to indicate the expected analysis result quality to the analyst. Moreover, the investigated QUALM modifiers lead to an increase in accuracy, e.g., of part-of-speech tagger and language identifier tools. In a qualitative discussion in this thesis, the positive effect of QUALM on a whole chain of analysis tools, i.e., an analysis pipeline, is shown.