Visualization of a multidimensional point cloud as a 3D swarm of avatars

TL;DR

The paper tackles the challenge of visualizing high-dimensional data by introducing avatar-based glyphs that semantically separate intuitive features (mapped to avatar visuals) from technical features (projected into spatial coordinates) within a $4$D data space. A filtering framework using a matrix $F\in\mathbb{R}^{(k+m)\times n}$ yields $X_{filtered}=F\cdot X$, with $X_{spatial}=F_{spatial}\cdot X$ and $X_{visual}=F_{visual}\cdot X$, followed by a view transform $X_{view}=V\cdot X_{spatial}$ for interactive 3D exploration. Implemented as an open-source plugin for the dpVision platform, the method supports slab-based filtering and real-time 3D projection of avatars derived from $X_{visual}$, enabling users to inspect structure and profiles simultaneously. Empirical demonstrations on synthetic politicians, synthetic soft drinks, and the Vinho Verde wine dataset show that explicit feature mapping enhances interpretability, facilitates outlier and archetype discovery, and provides immediate readouts of key attributes compared with conventional projection methods. This human-centered visualization approach offers practical advantages for analyzing complex multidimensional data and can be extended with more glyph types and automated variable-mapping strategies in future work.

Abstract

This paper proposes an innovative technique for representing multidimensional datasets using icons inspired by Chernoff faces. Our approach combines classical projection techniques with the explicit assignment of selected data dimensions to avatar (facial) features, leveraging the innate human ability to interpret facial traits. We introduce a semantic division of data dimensions into intuitive and technical categories, assigning the former to avatar features and projecting the latter into a four-dimensional (or higher) spatial embedding. The technique is implemented as a plugin for the open-source dpVision visualization platform, enabling users to interactively explore data in the form of a swarm of avatars whose spatial positions and visual features jointly encode various aspects of the dataset. Experimental results with synthetic test data and a 12-dimensional dataset of Portuguese Vinho Verde wines demonstrate that the proposed method enhances interpretability and facilitates the analysis of complex data structures.

Figures (13)

• Figure 1: Data dimensions vs display dimensions
• Figure 2: Conceptual diagram illustrating the separation of multidimensional data into spatial coordinates and avatar visual features through filtering and viewing transformations.
• Figure 3: Dialog window for assigning input dimensions to feature categories: spatial, visual, unnamed, or skipped. For spatial and visual categories, a specific output feature label can be selected.
• Figure 4: A cloud of points intersected by a slab, using a threshold of 1.5. Red points that lie outside the slab are discarded.
• Figure 5: Comparison of classical visualization methods for the synthetic politicians dataset. Top left: PCA scatter plot (first two components); Top right: t-SNE projection; Bottom left: parallel coordinates plot; Bottom right: Chernoff faces grid (4$\times$3) for all twelve politicians, colored and labeled by group. Each approach reveals different aspects of the data: scatter plots show grouping structure, parallel coordinates display multivariate profiles, and Chernoff faces summarize intuitive features---but none allows simultaneous interpretation of both structure and profile as directly as spatial avatars (see Fig. \ref{['fig:politicians_swarm']}).