Logo des Repositoriums

Auflistung nach Schlagwort "bias"

1 - 4 von 4
  • Textdokument
    Fairness in Regression -- Analysing a Job Candidates Ranking System
    (INFORMATIK 2022, 2022) Markert,Karla; Ahouzi,Afrae; Debus,Pascal
    Fairness is one of the pillars of any well-functioning society. Recent law-making in the EU regulates the machine-centered approach and thus increases the necessity for certifiable fairness approaches. In this paper, we adapt previous literature on certifiable fairness in classification systems to a regression model for simplified candidates ranking. This model serves as an illustration for an application that should work fairly even if built upon a biased data set. With our synthetic dataset we are able to analyse the challenges of different fairness notions. Although the fairness training manages to improve the certifiable individual fairness, some of the encoded bias remains. We discuss the challenges we faced, including the selection of suitable parameters and the trade off between accuracy and fairness. We hope to encourage more research into fairness improvement and certification, within and beyond group and individual fairness.
  • Konferenzbeitrag
    Mitigating Biases using an Additive Grade Point Model: Towards Trustworthy Curriculum Analytics Measures
    (21. Fachtagung Bildungstechnologien (DELFI), 2023) Baucks, Frederik; Wiskott, Laurenz
    Curriculum Analytics (CA) tries to improve degree program quality and learning experience by studying curriculum structure and student data. In particular, descriptive data measures (e.g., correlation-based curriculum graphs) are essential to monitor whether the learning process proceeds as intended. Therefore, identifying confounders and resulting biases and mitigating them should be critical to ensure reliable and fair results. Still, CA approaches often use raw student data without considering the influence of possible confounders such as student performance, course difficulty, workload, and time, which can lead to biased results. In this paper, we use an additive grade model to estimate these confounders and verify the validity and reliability of the estimates. Further, we mitigate the estimated confounders and investigate their impact on the CA measures course-to-course correlation and order benefit. Using data from 574 Computer Science Bachelor students, we show that these measures are significantly confounded and mislead to biased interpretations.
  • Konferenzbeitrag
    Peer-Reviewing and Submission Dynamics Around Top Software-Engineering Venues: A Juniors’ Perspective
    (Software Engineering 2023, 2023) Alchokr, Rand; Krüger, Jacob; Shakeel, Yusra; Saake, Gunter; Leich, Thomas
    Research is an intrinsically challenging process full of obstacles. However, these obstacles may be more dominant for a specific group of researchers (such as junior researchers) compared to others. It is the responsibility of the community to pay close attention to those groups that may be struggling for unfair reasons and provide necessary support. Junior researchers are of high importance to the scientific community, and are defined as young researchers who have recently started their research career[ Li19]. Despite their importance, juniors may face impediments when starting their career that hinder their activities and motivation. For instance, collaboration aspects and peer-reviewing models can play a role. Junior researchers without a high reputation (e.g., via their co-authors) may be negatively impacted by reputation biases, and thus could have even more problems with publishing and building their reputation independently. In our study, we investigate what challenges junior researchers perceive when submitting their work to software-engineering venues with a high reputation.
  • Textdokument
    Understanding Perceptual Bias in Machine Vision Systems
    (INFORMATIK 2020, 2021) Offert, Fabian; Bell, Peter
    Machine vision systems based on deep convolutional neural networks are increasingly utilized in digital humanities projects, particularly in the context of art-historical and audiovisual data. As research has shown, such systems are highly susceptible to bias. We propose that this is not only due to their reliance on biased datasets but also because their perceptual topology, their specific way of representing the visual world, gives rise to a new class of bias that we call perceptual bias. Perceptual bias, we argue, affects almost all currently available “off-the-shelf” machine vision systems, and is thus especially relevant for digital humanities applications, which often rely on such systems for hypothesis building. We evaluate the nature and scope of perceptual bias by means of a close reading of a visual analytics technique called “feature visualization” and propose to understand the development of critical visual analytics techniques as an important (digital) humanities challenge, situated at the interface of computer science and visual studies.