Logo des Repositoriums

Auflistung nach Schlagwort "Robustness"

1 - 4 von 4
  • Konferenzbeitrag
    Dealing with Hardware-related Disturbances in Organic Computing Systems
    (INFORMATIK 2019: 50 Jahre Gesellschaft für Informatik – Informatik für Gesellschaft (Workshop-Beiträge), 2019) Görlich-Bucher, Markus
    The ability to withstand disturbances while remaining functioning in a desired way is regarded as a crucial element in the field of Organic Computing. However, current approaches to self-healing and robustness fail in considering hardware-related breakdowns. These disturbances differ from software-sided disturbances in various aspects: They persist until being repaired, therefore their removal necessitates maintenance actions performed by human repair workers. Furthermore, they may be predicted to a certain degree. In this article, we formulate a problem statement and various requirements an OC system must fulfil in order to increase its robustness against hardware-related disturbances. Furthermore, we present a working plan for a PhD project concerning the investigation of several aspects of the previously motivated problem statement.
  • Textdokument
    GAFAI: Proposal of a Generalized Audit Framework for AI
    (INFORMATIK 2022, 2022) Markert,Thora; Langer,Fabian; Danos,Vasilios
    ML based AI applications are increasingly used in various fields and domains. Despite the enormous and promising capabilities of ML, the inherent lack of robustness, explainability and transparency limits the potential use cases of AI systems. In particular, within every safety or security critical area, such limitations require risk considerations and audits to be compliant with the prevailing safety and security demands. Unfortunately, existing standards and audit schemes do not completely cover the ML specific issues and lead to challenging or incomplete mapping of the ML functionality to the existing methodologies. Thus, we propose a generalized audit framework for ML based AI applications (GAFAI) as an anticipation and assistance to achieve auditability. This conceptual risk and requirement driven approach based on sets of generalized requirements and their corresponding application specific refinements as contributes to close the gaps in auditing AI.
  • Zeitschriftenartikel
    Robust visualization of trajectory data
    (it - Information Technology: Vol. 64, No. 4-5, 2022) Zhang, Ying; Klein, Karsten; Deussen, Oliver; Gutschlag, Theodor; Storandt,Sabine
    The analysis of movement trajectories plays a central role in many application areas, such as traffic management, sports analysis, and collective behavior research, where large and complex trajectory data sets are routinely collected these days. While automated analysis methods are available to extract characteristics of trajectories such as statistics on the geometry, movement patterns, and locations that might be associated with important events, human inspection is still required to interpret the results, derive parameters for the analysis, compare trajectories and patterns, and to further interpret the impact factors that influence trajectory shapes and their underlying movement processes. Every step in the acquisition and analysis pipeline might introduce artifacts or alterate trajectory features, which might bias the human interpretation or confound the automated analysis. Thus, visualization methods as well as the visualizations themselves need to take into account the corresponding factors in order to allow sound interpretation without adding or removing important trajectory features or putting a large strain on the analyst. In this paper, we provide an overview of the challenges arising in robust trajectory visualization tasks. We then discuss several methods that contribute to improved visualizations. In particular, we present practical algorithms for simplifying trajectory sets that take semantic and uncertainty information directly into account. Furthermore, we describe a complementary approach that allows to visualize the uncertainty along with the trajectories.
  • Zeitschriftenartikel
    Stance Detection Benchmark: How Robust is Your Stance Detection?
    (KI - Künstliche Intelligenz: Vol. 35, No. 0, 2021) Schiller, Benjamin; Daxenberger, Johannes; Gurevych, Iryna
    Stance detection (StD) aims to detect an author’s stance towards a certain topic and has become a key component in applications like fake news detection, claim validation, or argument search. However, while stance is easily detected by humans, machine learning (ML) models are clearly falling short of this task. Given the major differences in dataset sizes and framing of StD (e.g. number of classes and inputs), ML models trained on a single dataset usually generalize poorly to other domains. Hence, we introduce a StD benchmark that allows to compare ML models against a wide variety of heterogeneous StD datasets to evaluate them for generalizability and robustness. Moreover, the framework is designed for easy integration of new datasets and probing methods for robustness. Amongst several baseline models, we define a model that learns from all ten StD datasets of various domains in a multi-dataset learning (MDL) setting and present new state-of-the-art results on five of the datasets. Yet, the models still perform well below human capabilities and even simple perturbations of the original test samples (adversarial attacks) severely hurt the performance of MDL models. Deeper investigation suggests overfitting on dataset biases as the main reason for the decreased robustness. Our analysis emphasizes the need of focus on robustness and de-biasing strategies in multi-task learning approaches. To foster research on this important topic, we release the dataset splits, code, and fine-tuned weights.