Distance-preserving projections such as multi-dimensional scaling (MDS) or t-distributed stochastic neighborhood embedding (t-SNE) are popular methods to analyze high-dimensional data. The general idea is to project original data into a 2D layout by preserving the original similarities as well as possible. The analysis assumption is that close objects in the 2D projection correspond to similar objects in the high-dimensional space.
The visual patterns in 2D projections are often interpreted without questioning the quality of the projection. MDS, for example, optimizes the 2D layout by preserving all pair-wise similarities of the original data. Depending on the (dis-)similarity distribution, the MDS projection can reflect the original structures or not. The goal of this project is to develop methods for the quality assessment of such projections.
- Literature review for existing quality measures for projections.
- Development of quality criteria.
- Development of novel quality measures for distance-preserving projections.
- Development of a visual-interactive tool for the quality assessment of distance-preserving projections.
- Knowledge in information visualization.
- Scope: Bachelor/Master
- 6 Month Project, 3 Month Thesis (Bachelor) / 6 Month Thesis (Master)
- Start: immediately
- Probing Projections: Interaction Techniques for Interpreting Arrangements and Errors of Dimensionality Reductions [Stahnke et al., 2016]
- ProxiViz: an Interactive Visualization Technique to Overcome Multidimensional Scaling Artifacts [Heulot et al., 2012]
- IPCA: An interactive system for PCA-based visual analytics [Jeong et al., 2009]
- Data-driven Evaluation of Visual Quality Measures [Sedlmair and Aupetit, 2015]
- Data Visualization With Multidimensional Scaling [Buja et al., 2008]
- Modern multidimensional scaling [Borg and Groenen, 2005]
- Visualizing Data using t-SNE [Maaten and Hinton, 2008]