Visual-Interactive Machine Learning group conducts several research and application approaches that explicitly focus on Interactive Data Labeling. The central research question related to interactive data labeling is: which unlabeled instance should be labeled next to improve the quality of the training data most significantly? This is why the data labeling problem is also related to the research on strategies for an effective selection of candidate instances in the labeling process.
In general, GRIS investigates Interactive Data Labeling for classification, regression, and similarity search tasks. Recent publications span a wide range from basic research and theory to application-driven labeling approaches, involving domain experts from different domains. Examples include the assessment of patent well-being, labeling the similarity of soccer players, or training personal and user-centered music classifiers. Research in active learning, visualization and interaction design are combined to put the Visual-Interactive Labeling (VIAL) principle into practice which was proposed by Dr. Jürgen Bernard et al., the former head of the Interactive Machine Learning group.
Interactive Data Labeling explained
Interactive Data Labeling combines the traditional principle of labeling data (known in the domain) with novel interactive techniques (provided with the Machine Learing and Information Visualization domain). Active learning is a technique that is often applied in the machine learning context to label data more effectively. Interactive Data Labeling builds upon active learning, and conflates the approach with additional techniques from visual data analysis. As such, Interactive Data Labeling combines the strength of humans and machines in the tedious process of labeling (large) data sets. Visual Analytics
Visual Similarity Perception of Directed Acyclic Graphs: A Study on Influencing Factors and Similarity Judgment Strategies
VIAL: a unified process for visual interactive labeling
J Bernard, M Zeppelzauer, M Sedlmair, W Aigner
The Visual Computer, 1-19