211 results
Search Results
Now showing 1 - 10 of 211
Item DASH: Visual Analytics for Debiasing Image Classification via User-Driven Synthetic Data Augmentation(The Eurographics Association, 2022) Kwon, Bum Chul; Lee, Jungsoo; Chung, Chaeyeon; Lee, Nyoungwoo; Choi, Ho-Jin; Choo, Jaegul; Agus, Marco; Aigner, Wolfgang; Hoellt, ThomasImage classification models often learn to predict a class based on irrelevant co-occurrences between input features and an output class in training data. We call the unwanted correlations ''data biases,'' and the visual features causing data biases ''bias factors.'' It is challenging to identify and mitigate biases automatically without human intervention. Therefore, we conducted a design study to find a human-in-the-loop solution. First, we identified user tasks that capture the bias mitigation process for image classification models with three experts. Then, to support the tasks, we developed a visual analytics system called DASH that allows users to visually identify bias factors, to iteratively generate synthetic images using a state-of-the-art image-toimage translation model, and to supervise the model training process for improving the classification accuracy. Our quantitative evaluation and qualitative study with ten participants demonstrate the usefulness of DASH and provide lessons for future work.Item PerSleep: A Visual Analytics Approach for Performance Assessment of Sleep Staging Models(The Eurographics Association, 2021) Garcia Caballero, Humberto S.; Corvò, Alberto; Meulen, Fokke van; Fonseca, Pedro; Overeem, Sebasitaan; Wijk, Jarke J. van; Westenberg, Michel A.; Oeltze-Jafra, Steffen and Smit, Noeska N. and Sommer, Björn and Nieselt, Kay and Schultz, ThomasMachine learning is becoming increasingly popular in the medical domain. In the near future, clinicians expect predictive models to support daily tasks such as diagnosis and prognostic analysis. For this reason, it is utterly important to evaluate and compare the performance of such models so that clinicians can safely rely on them. In this paper, we focus on sleep staging wherein machine learning models can be used to automate or support sleep scoring. Evaluation of these models is complex because sleep is a natural process, which varies among patients. For adoption in clinical routine, it is important to understand how the models perform for different groups of patients. Moreover, models can be trained to recognize different characteristics in the data, and model developers need to understand why and how performance of the different models varies. To address these challenges, we present a visual analytics approach to evaluate the performance of predictive models on sleep staging and to help experts better understand these models with respect to patient data (e.g., conditions, medication, etc.). We illustrate the effectiveness of our approach by comparing multiple models trained on real-world sleep staging data with experts.Item Interactions for Seamlessly Coupled Exploration of High-Dimensional Images and Hierarchical Embeddings(The Eurographics Association, 2023) Vieth, Alexander; Lelieveldt, Boudewijn; Eisemann, Elmar; Vilanova, Anna; Höllt, Thomas; Guthe, Michael; Grosch, ThorstenHigh-dimensional images (i.e., with many attributes per pixel) are commonly acquired in many domains, such as geosciences or systems biology. The spatial and attribute information of such data are typically explored separately, e.g., by using coordinated views of an image representation and a low-dimensional embedding of the high-dimensional attribute data. Facing ever growing image data sets, hierarchical dimensionality reduction techniques lend themselves to overcome scalability issues. However, current embedding methods do not provide suitable interactions to reflect image space exploration. Specifically, it is not possible to adjust the level of detail in the embedding hierarchy to reflect changing level of detail in image space stemming from navigation such as zooming and panning. In this paper, we propose such a mapping from image navigation interactions to embedding space adjustments. We show how our mapping applies the "overview first, details-on-demand" characteristic inherent to image exploration in the high-dimensional attribute space. We compare our strategy with regular hierarchical embedding technique interactions and demonstrate the advantages of linking image and embedding interactions through a representative use case.Item KidCAD: An Interactive Cohort Analysis Dashboard of Patients with Chronic Kidney Diseases(The Eurographics Association, 2023) Höhn, Markus; Schwindt, Sarah; Hahn, Sara; Patyna, Sammy; Büttner, Stefan; Kohlhammer, Jörn; Angelini, Marco; El-Assady, MennatallahChronic Kidney Diseases (CKD) are a prominent health problem. With an ongoing process, CKD leads to impaired kindey function with decreased ability to filter the patients' blood, concluding in multiple complications, like heart disease and finally death. We developed a prototype to support nephrologists to gain an overview of their CKD patients. The prototype visualizes the patients in cohorts according to their pairwise similarity. The user can interactively modify the similarity by changing the underlying weights of the included features. The prototype was developed in response to the needs of physicians due to a context of use analysis. A qualitative user study shows the need and suitability of our new approach.Item Peeking at Visualization Research on Information Diffusion(The Eurographics Association, 2024) Usul, Mert; Arleo, Alessio; Kucher, Kostiantyn; Diehl, Alexandra; Gillmann, ChristinaDiffusion Processes are a widely researched topic of interest to different scientific domains. One of the most popular research directions is Information Diffusion, pertaining how information spreads over a tightly connected network. From the modeling perspective, many different approaches are known in the literature; however, in the visualization community, this still represents an under-investigated problem. In this work, we present a succinct overview of the current state-of-the-art in Visual Analytics techniques employed in representing and understanding diffusion processes happening over networks. We consider different application domains and introduce a taxonomy that categorizes and provides structure to our selection of papers, fostering further research in the field of Visual Analytics of Information Diffusion processes.Item Visual-Interactive Preprocessing of Multivariate Time Series Data(The Eurographics Association and John Wiley & Sons Ltd., 2019) Bernard, Jürgen; Hutter, Marco; Reinemuth, Heiko; Pfeifer, Hendrik; Bors, Christian; Kohlhammer, Jörn; Gleicher, Michael and Viola, Ivan and Leitte, HeikePre-processing is a prerequisite to conduct effective and efficient downstream data analysis. Pre-processing pipelines often require multiple routines to address data quality challenges and to bring the data into a usable form. For both the construction and the refinement of pre-processing pipelines, human-in-the-loop approaches are highly beneficial. This particularly applies to multivariate time series, a complex data type with multiple values developing over time. Due to the high specificity of this domain, it has not been subject to in-depth research in visual analytics. We present a visual-interactive approach for preprocessing multivariate time series data with the following aspects. Our approach supports analysts to carry out six core analysis tasks related to pre-processing of multivariate time series. To support these tasks, we identify requirements to baseline toolkits that may help practitioners in their choice. We characterize the space of visualization designs for uncertainty-aware pre-processing and justify our decisions. Two usage scenarios demonstrate applicability of our approach, design choices, and uncertainty visualizations for the six analysis tasks. This work is one step towards strengthening the visual analytics support for data pre-processing in general and for uncertainty-aware pre-processing of multivariate time series in particular.Item Immersive Analytics of Heterogeneous Biological Data Informed through Need-finding Interviews(The Eurographics Association, 2021) Ripken, Christine; Tusk, Sebastian; Tominski, Christian; Vrotsou, Katerina and Bernard, JürgenThe goal of this work is to improve existing biological analysis processes by means of immersive analytics. In a first step, we conducted need-finding interviews with 12 expert biologists to understand the limits of current practices and identify the requirements for an enhanced immersive analysis. Based on the gained insights, a novel immersive analytics solution is being developed that enables biologists to explore highly interrelated biological data, including genomes, transcriptomes, and phenomes. We use an abstract tabular representation of heterogeneous data projected onto a curved virtual wall. Several visual and interactive mechanisms are offered to allow biologists to get an overview of large data, to access details and additional information on the fly, to compare selected parts of the data, and to navigate up to about 5 million data values in real-time. Although a formal user evaluation is still pending, initial feedback indicates that our solution can be useful to expert biologists.Item Guidance or No Guidance? A Decision Tree Can Help(The Eurographics Association, 2018) Ceneda, Davide; Gschwandtner, Theresia; May, Thorsten; Miksch, Silvia; Streit, Marc; Tominski, Christian; Christian Tominski and Tatiana von LandesbergerGuidance methods have the potential of bringing considerable benefits to Visual Analytics (VA), alleviating the burden on the user and allowing a positive analysis outcome. However, the boundary between conventional VA approaches and guidance is not sharply defined. As a consequence, framing existing guidance methods is complicated and the development of new approaches is also compromised. In this paper, we try to bring these concepts in order, defining clearer boundaries between guidance and no-guidance. We summarize our findings in form of a decision tree that allows scientists and designers to easily frame their solutions. Finally, we demonstrate the usefulness of our findings by applying our guideline to a set of published approaches.Item Supporting Visual Parameter Analysis of Time Series Segmentation with Correlation Calculations(The Eurographics Association, 2018) Eichner, Christian; Schumann, Heidrun; Tominski, Christian; Anna Puig and Renata RaidouParameter analysis can be used to find out how individual parameters influence the output of an algorithm. We aim to support the visual parameter analysis of algorithms for the segmentation of time series. To this end, we automatically search for correlations between parameters and the segmentation outputs. Correlations are not only determined globally, but also locally within parameter subspaces. Calculated correlations are used to visually emphasize parameter and value ranges with high influence on the segmentation. By interactive exploration, the analyst can study the multidimensional parameter space in depth.Item Visual Predictive Analytics using iFuseML(The Eurographics Association, 2018) Sehgal, Gunjan; Rawat, Mrinal; Gupta, Bindu; Gupta, Garima; Sharma, Geetika; Shroff, Gautam; Christian Tominski and Tatiana von LandesbergerSolving a predictive analytics problem involves multiple machine learning tasks in a workflow. Directing such workflows efficiently requires an understanding of data so as to identify and handle missing values and outliers, compute feature correlations and to select appropriate models and hyper-parameters. While traditional machine learning techniques are capable of handling these challenges to a certain extent, visual analysis of data and results at each stage can significantly assist in optimal processing of the workflow. We present iFuseML , a visual interactive framework to support analysts in machine learning workflows via insightful data visualizations as well as natural language interfaces where appropriate. Our platform lets the user intuitively search and explore datasets, join relevant datasets using natural language queries, detect and visualize multidimensional outliers and explore feature relationships using Bayesian coordinated views. We also demonstrate how visualization assists in comparing prediction errors to guide ensemble models so as to generate more accurate predictions. We illustrate our framework using a house price dataset from Kaggle, where using iFuseML simplified the machine learning workflow and helped improve the resulting prediction accuracy.