Optimization and Machine Learning for Visualization
headed by Dr. Steffen Frey
Optimization and machine learning solve various types of visualization-related problems in a data-driven way. Optimization can be used to reduce and enrich the data, e.g., by identifying representative time steps from a time series, or quantifying similarity between data partitions. Visualization techniques also typically expose a variety of parameters influencing what can be seen (e.g., via the camera position), and automatic adaptation regarding a specified objective can guide user exploration. Optimization is also useful for dynamically adapting a visualization system at runtime, e.g., to equalize load between different machines or to yield a good balance between image quality and frame rate.
Machine learning techniques are capable of learning the structure of the data which can be exploited for visual analysis. Instead of requiring detailed explicit feature specification, users may specify aspects of interest from the data directly. This also allows for an automatic analysis of the full data, e.g., checking for events matching none of the common patterns. In addition, the transformation of data into different representation spaces can effectively be employed for similarity search and visual pattern analysis. Finally, performance prediction models provide the basis for (dynamically) adapting visualization systems.