Search Results

Now showing 1 - 4 of 4
  • Item
    Assessing the Reliability of Integrated Gradients-Based Saliency Maps for 3D Point Cloud Semantic Segmentation Models
    (The Eurographics Association, 2024) Ciprián-Sánchez, Jorge F.; Burmeister, Josafat-Mattias; Cech, Tim; Richter, Rico; Döllner, Jürgen; Hunter, David; Slingsby, Aidan
    Deep learning models achieve high accuracy in the semantic segmentation of 3D point clouds; however, it is challenging to discern which patterns a model has learned and how it derives its output from the input. Recently, the Integrated Gradients method has been adopted to explain semantic segmentation models for 3D point clouds. This method can be used to generate saliency maps that visualize the contribution of input points to a particular model output. However, there is a lack of quantitative evaluation of the reliability of the generated saliency maps and the influence of the baseline selection (a central component of Integrated Gradients) on the method's results. In this paper, we quantitatively evaluate the reliability of saliency maps generated by the Integrated Gradients method for a 3D point cloud semantic segmentation model through well-known sanity checks from the image domain that we adapt to 3D point cloud segmentation. We perform these sanity checks for three different baselines to further evaluate the stability of the generated saliency maps concerning the baseline choice. Our results indicate that the Integrated Gradients method is sensitive to both the parameters of the model and training labels, unstable concerning the choice of baseline, and that, although it can identify points with high contributions to the model output, it fails to identify correctly if such contributions are positive or negative. Finally, we propose an averaging approach to aggregate the results of points that receive multiple scores from Integrated Gradients during the segmentation process and show that it produces saliency maps that better reflect high-contribution input points than previous approaches.
  • Item
    Reflections on the Evolution of the BookTracker Visualization Platform
    (The Eurographics Association, 2024) Xing, Yiwen; Dondi, Cristina; Borgo, Rita; Abdul-Rahman, Alfie; Hunter, David; Slingsby, Aidan
    Understanding the trade data of historical books is crucial for researchers investigating the distribution and provenance of Incunabula (books printed between 1450 and 1500). We incrementally developed BookTracker, a platform featuring multiple visualization and visual analytics applications to support these research efforts. This platform leverages data from the Material Evidence in Incunabula (MEI) database, which meticulously records detailed information on the provenance, ownership, and use of 15th-century printed books. BookTracker began with a focus on providing visualization and visual analytical solutions to effectively present each book provenance's chronological and geographical information. Through three years of collaborative work with domain experts, we continually explored the Material Evidence in Incunabula (MEI) data and discovered more possibilities for visualization to represent this rich information. Gradually, a suite of specialized visualization tools for specific analytical purposes was developed, including DanteSearchVis, DanteExploreVis, KURF2022, KURF2023, and OwnershipTracker. These tools now comprise the BookTracker platform, which has evolved to explore various features and aspects of the data. This paper details the evolution of BookTracker's design and development alongside domain experts, highlighting the reflections and lessons learned from its application in various research projects. We discuss this long-term collaborative visualization project, hoping to offer our experience as a case study for similar research in the future.
  • Item
    Multi-level Visualization for Exploration of Structures in Missing Data
    (The Eurographics Association, 2024) Alsufyani, Sarah; Forshaw, Matthew; Din, Silvia Del; Yarnall, Alison; Rochester, Lynn; Fernstad, Sara Johansson; Hunter, David; Slingsby, Aidan
    Missing data refers to the absence of a value in the dataset where it was expected to be present. This absence is common across various fields. It can be caused by a range of factors in the data collection process, and may severely impact analysis through unreliable or biased results. Missing data visualization provides an effective approach to exploring the missing data, recognizing the missingness patterns and structures, and determining optimal solutions through interactive visual interfaces. This paper presents a visualization prototype that incorporates two novel techniques, the MissVisG glyph and the MissVis plot, to support the exploration of missing values in data. The visualization provides an overview of missing values, and helps identify patterns in the data to guide users in selecting appropriate methods for dealing with the missingness. A multi-step evaluation process is utilized to assess and ensure the usability and effectiveness of the visualization.
  • Item
    Visualizing Complex Data Decisions: Design Study for Ethical Factors in AI Clinical Decision Support Systems
    (The Eurographics Association, 2024) Surodina, Svitlana; Volkova, Daria; Abdul-Rahman, Alfie; Borgo, Rita; Hunter, David; Slingsby, Aidan
    Despite the proliferation of Artificial Intelligence (AI) technologies, their uptake in clinical settings has been lacking progress due to complexities of sociotechnical factors and intricacies of decision-making. Fairness and bias of predictive models, ethics and quality of training data, and corresponding compliance requirements become especially pressing while remaining fuzzy and implicit for various stakeholders who make the decisions. We present learnings and future directions from a design study with domain experts and propose a novel approach to encoding and collaborative reasoning on complex requirements for AI-Empowered Clinical Decision Support System (AI-CDSS) design based on Knowledge Graph (KG) representation. The insights will be useful to the community of visualization researchers who work on ethical AI-CDSS design and conduct design studies with clinical partners.