Research - Overview
Research at Computer Science 9 aims at data mining and database technology for multimedia and spatio-temporal databases. In many applications, more and more digital information is generated and needs to be stored, processed, explored and analyzed. In our projects, we consider data from mechanical engineering, civil engineering, mobile communication, medical diagnostics, molecular biology, environmental sciences and multimedia scenarios in general.
Data Analysis and Knowledge Extraction
Increasingly large data resources require automatic techniques for gaining knowledge which means extracting interesting, unknown patterns from the data. In scenarios with many attributes or with noise, these patterns are typically hidden in subspaces of the data and do not show up in the full dimensional space; therefore, we develop new data mining techniques including subspace clustering or outlier detection. For data streams which constitute endless data resources, we develop specialized algorithms that can handle both the infinite amount of data and the limited and often varying amount of time available between two stream data items. Typical tasks are clustering of streaming data or classification on data streams.
- SteerSCiVA: Steerable Subspace Clustering for Visual Analytics
- SFB 686
- MOA - Massive Online Analysis
- Anytime Stream Mining
- Research cluster UMIC
- Subspace Mining for High Dimensional Data
- Outlier Ranking for High Dimensional Data
- OpenSubspace Framework
- Clustering in Attributed Graphs
- NISIS - Nature Inspired Intelligent Systems
Exploration of Multimedia Databases
Today's scientific, commercial, and entertainment applications produce large amount of multimedia data. In order to get insight into these data, new exploration models are needed. Aiming at reflecting the user's perception, we develop new content-based similarity models based on adaptable distance functions, novel relevance feedback concepts that allow exploring comprehensive multimedia databases, and new interactive visualization techniques to make the exploration process more accessible and more intuitive for the user. These models are computationally expensive and we develop new methods for efficient query processing.
Fast Access to Complex Data
Tasks as similarity search or data analysis demand for fast access to complex data. For this, we develop new techniques, as for example index structures, approximation methods and efficient query processing algorithms for complex objects including high dimensional data, time series, or interval data.