Browsing by Author "Kwon, Bum Chul"

Now showing 1 - 2 of 2
  • Thumbnail Image
    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, Thomas
    Image 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.
  • Thumbnail Image
    Item
    VATUN: Visual Analytics for Testing and Understanding Convolutional Neural Networks
    (The Eurographics Association, 2021) Park, Cheonbok; Yang, Soyoung; Na, Inyoup; Chung, Sunghyo; Shin, Sungbok; Kwon, Bum Chul; Park, Deokgun; Choo, Jaegul; Agus, Marco and Garth, Christoph and Kerren, Andreas
    Convolutional neural networks (CNNs) are popularly used in a wide range of applications, such as computer vision, natural language processing, and human-computer interaction. However, testing and understanding a trained model is difficult and very time-consuming. This is because their inner mechanisms are often considered as a 'black-box' due to difficulty in understanding the causal relationships between processes and results. To help the testing and understanding of such models, we present a user-interactive visual analytics system, VATUN, to analyze a CNN-based image classification model. Users can accomplish the following four tasks in our integrated system: (1) detect data instances in which the model confuses classification, (2) compare outcomes of the model by manipulating the conditions of the image, (3) understand reasons for the prediction of the model by highlighting highly influential parts from the image, and (4) analyze the overall what-if scenarios when augmenting the instances for each class. Moreover, by combining multiple techniques, our system lets users analyze behavior of the model from various perspectives. We conduct a user study of an image classification scenario with three domain experts. Our study will contribute to reducing the time cost for testing and understanding the CNN-based models in several industrial areas.