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 Load-balancing for scalable simulations with large particle numbers(2021) Hirschmann, Steffen; Pflüger, Dirk (Prof. Dr.)Item Open Access Metrics and algorithms for locally fair and accurate classifications using ensembles(2022) Lässig, Nico; Oppold, Sarah; Herschel, MelanieTo obtain accurate predictions of classifiers, model ensembles comprising multiple trained machine learning models are nowadays used. In particular, dynamic model ensembles pick the most accurate model for each query object, by applying the model that performed best on similar data. Dynamic model ensembles may however suffer, similarly to single machine learning models, from bias, which can eventually lead to unfair treatment of certain groups of a general population. To mitigate unfair classification, recent work has thus proposed fair model ensembles , that instead of focusing (solely) on accuracy also optimize global fairness . While such global fairness globally minimizes bias, imbalances may persist in different regions of the data, e.g., caused by some local bias maxima leading to local unfairness . Therefore, we extend our previous work by including a framework that bridges the gap between dynamic model ensembles and fair model ensembles. More precisely, we investigate the problem of devising locally fair and accurate dynamic model ensembles, which ultimately optimize for equal opportunity of similar subjects. We propose a general framework to perform this task and present several algorithms implementing the framework components. In this paper we also present a runtime-efficient framework adaptation that keeps the quality of the results on a similar level. Furthermore, new fairness metrics are presented as well as detailed informations about necessary data preparations. Our evaluation of the framework implementations and metrics shows that our approach outperforms the state-of-the art for different types and degrees of bias present in training data in terms of both local and global fairness, while reaching comparable accuracy.Item Open Access Performance-oriented communication concepts for networked control systems(2022) Carabelli, Ben W.; Rothermel, Kurt (Prof. Dr. rer. nat. Dr. h. c.)Networked control systems (NCS) integrate sensors, actuators, and digital controllers using a communication network in order to control physical processes. They can be found in diverse application areas, including automotive and aircraft systems, smart homes, and smart manufacturing systems in the context of Industry 4.0. Because control systems have demanding Quality of Service (QoS) requirements, the provisioning of appropriate communication services for NCS is a challenge. Moreover, the trend of steadily increasing digitization in many fields will likely lead to control applications with more complex system integration, especially in large-scale systems such as smart grids and smart cities. The proliferation of NCS in such an environment clearly depends on strong methods for integrating communication and control. However, there currently remains a gap between these two domains. On the one hand, the control-theoretic design and analysis methods for NCS have been based on simplistic and abstract network connection models. On the other hand, communication networks are optimized for conventional performance metrics such as throughput and latency, which do not readily translate into application specific Quality of Control (QoC) metrics. The goal of this thesis is to provide performance-oriented concepts for the design of communication services for NCS. In particular, methods for scheduling and routing the traffic of NCS and increasing their reliability through replication are developed on the basis of integrated models that capture the relationship between control-relevant characteristics of communication services and the methods that are used to provide those communication services in the network. This thesis makes the following contributions. First, we address the problem of optimally arbitrating limited communication bandwidth for a group of NCS in a shared network by designing a performance-aware dynamic priority scheduler. The resulting first scheduling policy provides asymptotic stability guarantees for each NCS and performance bounds on the joint QoC. While it is efficient to implement on the data link layer with stateless priority queueing, it requires a large optimization problem comprising all NCS to be solved initially for determining scheduler parameters. To increase the scalability, we therefore relax the scheduling problem by separating the NCS traffic into deterministic transmissions with real-time guarantees and opportunistic traffic used for QoC optimization. The resulting second scheduling policy imposes no QoS constraints on opportunistic traffic, yields less conservative stability guarantees, and allows scheduler parameters to be calculated for each NCS separately and thus much more efficiently. Second, we address the problem of optimally routing NCS traffic in networks with random latency distributions by designing a cross-layer communication service for stochastic NCS. The routing algorithm exploits trade-offs between delay and in-time arrival probabilities to find a route that provides a predefined level of QoC while minimizing network load. Third, we address the problem of active replication for controllers in order to increase the reliability of NCS subject to crash failures and message loss. While existing replication schemes for real-time systems focus only on ensuring that no conflicting values are sent to actuators, we develop stronger consistency concepts that provide replication transparency for control systems. We present a corresponding replication management protocol that achieves high availability and low latency at low message cost, and evaluate it using physical experiments.Item Open Access A preCICE-FMI Runner to couple controller models to PDEs(2023) Willeke, LeonardPartitioned simulation or co-simulation allows to simulate complex systems by breaking them into smaller, independent subsystems. The Functional Mock-Up Interface FMI enables co-simulation by defining a framework for simulation models. Models adhering to the standard interface (FMUs) are executed and coupled by an importer. This framework approach works well for models based on ODEs and DAEs but reaches its limits for models based on PDEs. Such models require sophisticated, legacy software packages not compatible with the FMI standard. However, only PDE-based models are able to accurately simulate many physical aspects important in engineering like heat transfer or Fluid-Structure interactions. A possible solution to this problem is the open-source coupling library preCICE. preCICE couples PDE-based simulation programs in a black-box fashion to achieve partitioned multi-physics simulations. The coupling of the FMI standard to preCICE would allow the co-simulation of FMI models with the more than 20 simulation programs in the preCICE ecosystem. This thesis is focused on the development of a preCICE-FMI Runner software to couple FMUs with preCICE. The Runner serves as an importer to execute the FMU and steer the simulation. Additionally, it calls the preCICE library to communicate and coordinate with other solvers. The scope is not to develop a general Runner software, but to couple FMUs that contain control algorithms with PDE-based models as a first step. The software is written in Python and relies on the Python package FMPy as well as the preCICE Python bindings. Two test cases show the functionality of the preCICE-FMI Runner. The coupling of ODE-based models with FMUs matches the results of a pure Python implementation with an accuracy of 10 −4 . The coupling of a PDE-based model to a controller FMU proofs the working principle, although the results could not be tested against other implementations. The scope of the implemented abilities restricts the possible simulation scenarios, but does not prohibit a general use for coupling scenarios beyond control applications.Item Open Access AssistML : an approach to manage, recommend and reuse ML solutions(2023) Villanueva Zacarias, Alejandro Gabriel; Reimann, Peter; Weber, Christian; Mitschang, BernhardThe adoption of machine learning (ML) in organizations is characterized by the use of multiple ML software components. When building ML systems out of these software components, citizen data scientists face practical requirements which go beyond the known challenges of ML, e. g., data engineering or parameter optimization. They are expected to quickly identify ML system options that strike a suitable trade-off across multiple performance criteria. These options also need to be understandable for non-technical users. Addressing these practical requirements represents a problem for citizen data scientists with limited ML experience. This calls for a concept to help them identify suitable ML software combinations. Related work, e. g., AutoML systems, are not responsive enough or cannot balance different performance criteria. This paper explains how AssistML, a novel concept to recommend ML solutions, i. e., software systems with ML models, can be used as an alternative for predictive use cases. Our concept collects and preprocesses metadata of existing ML solutions to quickly identify the ML solutions that can be reused in a new use case. We implement AssistML and evaluate it with two exemplary use cases. Results show that AssistML can recommend ML solutions in line with users’ performance preferences in seconds. Compared to AutoML, AssistML offers citizen data scientists simpler, intuitively explained ML solutions in considerably less time. Moreover, these solutions perform similarly or even better than AutoML models.