Semantische Suche

In dynamical systems, like gas turbines, transition phases occur after a change in the input control signal. The domain knowledge about the steepness of these transitions can potentially help with the modeling of such systems using the data science approaches. There already exist TGDS approaches that use the information about the limits of the values. However it is currently not clear how to incorporate the information about the steepness of the transitions with them.

In this thesis, we develop three different TGDS approaches to include these transition constraints in recurrent neural networks (RNNs) to improve the modeling of input-output behavior of dynamical systems. We evaluate the approaches on synthetic and real time series data by varying data availability and different degrees of steepness. We conclude that the TGDS approaches are especially helpful for flat transitions and provide a guideline on how to use the available transition constraints in real world problems. Finally, we discuss the required degree of domain knowledge and intellectual implementation effort of each approach.

While some approaches use additional information about the current state of the environment, bandit algorithms tend to ignore domain knowledge that can’t be extracted from data. It is not clear how to incorporate domain knowledge into bandit algorithms and how much improvement this yields.

In this masters thesis we propose two ways to augment bandit algorithms with domain knowledge: a push approach, which influences the distribution of arms to deal with non-stationarity as well as a group approach, which propagates feedback between similar arms. We conduct synthetic and real world experiments to examine the usefulness of our approaches. Additionally we evaluate the effect of incomplete and incorrect domain knowledge. We show that the group approach helps to reduce exploration time, especially for small number of iterations and plays, and that the push approach outperforms contextual and non-contextual baselines for large context spaces.

In unserer Arbeit stellen wir einen Ansatz vor, der aus einer gegebenen repräsentativen Benchmark eine kleinere Teilmenge wählt, die als repräsentative Benchmark dienen soll. Wir definieren dabei, dass eine Benchmark repräsentativ ist, wenn der Graph der Laufzeiten ein festgelegtes Ähnlichkeitsmaß gegenüber der ursprünglichen Benchmark überschreitet. Wir haben hierbei einen BeamSearch-Algorithmus erforscht. Am Ende stellen wir allerdings fest, dass eine zufällige Auswahl besser ist und eine zufällige Auswahl von 10 % der Instanzen ausreicht, um eine repräsentative Benchmark zu liefern.

We extend an existing data flow diagram approach with our concept of trust. Our approach of adding knowledge to a software architecture model and providing a way to analyze model instances for access control violations shall enable software architects to increase the quality of models and further verify access control requirements under uncertainty. We evaluate the applicability based on the availability, the accuracy and the scalability regarding the execution time.

For many explainers this proposed reasoning gives us more information about the inner workings of the model or even about the training data. Since data privacy is becoming an important issue the question arises whether explainers can leak private data. It is unclear what private data can be obtained from different kinds of explanation. In this thesis I adapt three privacy attacks in machine learning to the field of XAI: model extraction, membership inference and training data extraction. The different kinds of explainers are sorted into these categories argumentatively and I present specific use cases how an attacker can obtain private data from an explanation. I demonstrate membership inference and training data extraction for two specific explainers in experiments. Thus, privacy can be breached with the help of explainers.

To close the gap between the development and having up-to-date performance models, the Continuous Integration of Performance Models (CIPM) approach has been proposed. It incorporates automatically executed activities into a Continuous Integration pipeline and is realized with Vitruvius combining Java and the PCM. As a consequence, changes from a commit are extracted to incrementally update the models in the VSUM. To estimate the resource demand in the PCM, the CIPM approach adaptively instruments and monitors the source code.

In previous work, parts of the CIPM pipeline were prototypically implemented and partly evaluated with artificial projects. A pipeline combining the incremental model update and the adaptive instrumentation is absent. Therefore, this thesis presents the combined pipeline adapting and extending the existing implementations. The evaluation is performed with the TeaStore and indicates the correct model update and instrumentation. Nevertheless, there is a gap towards the calibration pipeline.

The goal of this Master’s thesis is to develop and evaluate a framework to efficiently and effectively detect “when” and “where” concept drift occurs in high-dimensional data streams. We introduce stream autoencoder windowing (SAW), an approach based on the online training of an autoencoder, while monitoring its reconstruction error via a sliding window of adaptive size. We will evaluate the performance of our method against synthetic data, in which the characteristics of drifts are known. We then show how our method improves the accuracy of existing classifiers for predictive systems compared to benchmarks on real data streams.

Logs are similar to natural languages. Recent deep learning algorithm Transformer Neural Network has shown outstanding performance in Natural Language Processing (NLP) tasks. Based on these, we adapt Transformer Neural Network to detect outliers from applications logs In an unsupervised way. We compared our algorithm against state-of-the-art log outlier detection algorithms on three widely used benchmark datasets. Our algorithm outperformed state-of-the-art log outlier detection algorithms.

Despite the existing work in the healthcare context, making general observations about the impact of domain knowledge is difficult, as different publications use different knowledge types, prediction tasks and model architectures. It further remains unclear if the findings in healthcare are transferable to other use-cases, as well as how much intellectual effort this requires.

With this Thesis we introduce DomainML, a modular framework to evaluate the impact of domain knowledge on different data science tasks. We demonstrate the transferability and flexibility of DomainML by applying the concepts from healthcare to a cloud system monitoring. We then observe how domain knowledge impacts the model’s prediction performance across both domains, and suggest how DomainML could further be used to refine both the given domain knowledge as well as the quality of the underlying dataset.

Design Space Exploration can be a resource intensive process. An architecture candidate may feature certain properties which disqualify it from consideration as an optimal candidate, regardless of its quality metrics. An example for this would be confidentiality violations in data flows introduced by certain components or combinations of components in the architecture. If these properties can be identified early, quality evaluation can be skipped and the candidate discarded, saving resources.

Currently, analyses for identifying such properties are performed disjunct from the design space exploration process. Optimal candidates are determined first, and analyses are then applied to singular architecture candidates. Our approach augments the PerOpteryx design space exploration pipeline with an additional architecture candidate filter stage, which allows existing generic candidate analyses to be integrated into the DSE process. This enables automatic execution of analyses on architecture candidates during DSE, and early discarding of unwanted candidates before quality evaluation takes place.

We use our filter stage to perform data flow confidentiality analyses on architecture candidates, and further provide a set of example analyses that can be used with the filter. We evaluate our approach by running PerOpteryx on case studies with our filter enabled. Our results indicate that the filter stage works as expected, able to analyze architecture candidates and skip quality evaluation for unwanted candidates.

Die in der Arbeit entwickelte Erweiterung des Palladio-Konzeptes ermöglicht es, diese Analyse schon zur Entwurfszeit anhand eines Modells durchzuführen. Mithilfe der Erweiterung kann analysiert werden, wie oft und in welchem Verhältnis ein Fehler aufgetreten ist, welche Fehlervorkommen miteinander korrelieren und wie schwerwiegend die Auswirkungen der aufgetretenen Fehler für den Systemkontext waren. Neben der Analyse der Fehlerpropagation ermöglicht die Erweiterung die Modellierung von Systemen, die auf das Vorkommen eines Fehlers im Sinne einer Rekonfiguration reagieren. Das Konzept wurde anhand eines sicherheitskritischen Systems aus der Domäne der autonomen Fahrzeuge validiert.

In meinem Ansatz wird ein Schema zur Einordnung entwickelt, das sich an aktuelle Literatur anlehnt und drei grundsätzliche Arten von Entscheidungen unterscheidet: Existenzentscheidungen, Eigenschaftenentscheidungen und Umgebungsentscheidungen. Zur Evaluation wurden Open-Source-Softwareprojekte mit natürlichsprachiger Softwarearchitekturdokumentationen betrachtet und iterativ überprüft, wo das aktuelle Schema verbessert werden kann. Zum Schluss wird vorgestellt, welche der Entscheidungsklassen sich im Palladio Component Model abbilden lassen.

In this thesis we will analyze and visualize semantics from massive document directories. The results will be displayed using the arXiv corpus, which contains domain-specific (computer science) papers of the past thirty years. The analysis will illustrate and give insight about past trends of document directories and how their relationships evolve over time.

When working with large data sets, in many situations one has to deals with a large set data from a single class and only few negative examples from other classes. Learning classifiers, which can assign data points to one of the groups, is known as one-class classification (OCC) or outlier detection.

The objective of this thesis is to develop and evaluate an active learning process to train an OCC. The process uses domain knowledge to reasonably adopt a prior distribution. Knowing that prior distribution, query strategies will be evaluated, which consider the certainty, more detailed the uncertainty, of the estimated class membership scorings. The integration of the prior distribution and the estimation of uncertainty, will be modeled using a gaussian process.

Verfahren der künstlichen Intelligenz auf der Grundlage neuronaler Netzwerke können genutzt werden, um viele der besonders aufwändigen Aufgaben schneller oder sogar besser zu lösen als herkömmliche Methoden. In dieser Arbeit wird ein neuartiges neuronales Netzwerk auf seine Fähigkeit hin untersucht, eine Software allein anhand der Pixeldaten ihrer Benutzeroberfläche zu testen. Des Weiteren wird ein Framework entwickelt, welches mithilfe von leistungsfähigen GPUs den Trainingsvorgang deutlich beschleunigen kann.

The recently introduced Monte Carlo Dependency Estimation (MCDE) framework defines the dependency between a set of variables as the expected value of a stochastic process performed on them. In practice, this expected value is approximated with an estimator which iteratively performs a set of Monte Carlo simulations. In this thesis, we propose several alternative estimators to approximate this expected value. They function in a more dynamic way and also leverage information from previous approximation iterations. Using both probability theory and experiments, we show that our new estimators converge much faster than the original one.