The Visual-Interactive Machine Learning Group

The Visual-Interactive Machine Learning Group

Motivation

The demand for effective machine learning solutions in research and application areas has never been greater than today. With the increasing number of data scientists involved machine learning, traditional problems and challenges of the machine learning process now also affect broader groups than ever before. Traditional machine learning problems often related to the blackbox-nature of machine learning models. In most cases, large parts of machine learning processes (data collection, preprocessing, feature selection, model building, model validation) can hardly be observed by users in a visual way. As a consequence, users are often left alone with trust-building, (visual) validation, and the informed refinement of machine learning models. In addition, we underline the problem that machine learning is often a model-centered rather than a user-centered process; and the involvement of end users in the process is often neglected.

Research

The Visual-Interactive Machine Learning group at GRIS addresses these problems with a clear focus on visualization, interaction, and human factors. We follow the overall goal to enable data scientists AND end users (without in-depth knowledge in data science) to participate in the machine learning process. Our approach is to open the blackbox of the machine learning process and make model building and refinement a visual-interactive endeavor. Visual-interactive interfaces based on the theories and applications in information visualization and visual analytics can be conflated with best practices in traditional machine learning. Given that, the visual-interactive machine learning process fosters the iterative nature of the process, including faster, target-oriented, and incremental learning cycles. Finally, we investigate human factors in the machine learning process with the goal to design better visual-interactive interfaces, leading to a more human-centered machine learning process.

Teaching and Student Projects

The lectures about Visual Analytics and User-Centered Design in Visual Computing have a natural connecting point to Visual-Interactive Machine Learning. In addition, several practical course (such as Visual Computing Lab) build a valuable basis for students to encounter Visual-Interactive Machine Learning principles.

References

  • VIAL: a unified process for visual interactive labeling. J Bernard, M Zeppelzauer, M Sedlmair, W Aigner. The Visual Computer, 1-19
  • Towards User‐Centered Active Learning Algorithms. J Bernard, M Zeppelzauer, M Lehmann, M Müller, M Sedlmair. Computer Graphics Forum 37 (3), 121-132
  • Comparing visual-interactive labeling with active learning: An experimental study. J Bernard, M Hutter, M Zeppelzauer, D Fellner, M Sedlmair. IEEE transactions on visualization and computer graphics 24 (1), 298-308