Dust & Magnet · Feature-space exploration

Explore feature spaces with analytical magnets.

An interpretable, particle-based approach to interactive, human-in-the-loop exploration of high-dimensional feature spaces.

Represent feature vectors as particles. Define attraction and repulsion criteria that stay tied to semantically meaningful dimensions.

Start exploring

The research model

From feature vectors
to explicit hypotheses.

High-dimensional datasets describe objects through many features, making relationships, similarities, and outliers difficult to interpret. Dust & Magnet visualization represents these objects as particles, with magnets attracting or repelling them according to feature values.

Analytical magnets extend this metaphor with criteria that can be combined interactively, including logical AND and OR, similarity, variance, and selected groups of features. Moving magnets and adjusting their influence lets analysts explore hypotheses while keeping the underlying dimensions and criteria inspectable. Particle placement combines analytical barycentric targets with collision-aware local relaxation to reduce overlap, providing an interpretable, human-in-the-loop approach to exploring feature spaces.

Related publications

Research behind the prototypes

  1. iV ’26 · IEEEIn press

    Feature Space Exploration by Particle-Based Visualization Controlled by Analytical Magnets

    Jürgen Döllner and Willy Scheibel

    Proceedings of the 30th International Conference Information Visualisation, 2026.

  2. Web3D ’26 · ACMIn press

    A WebGPU Processing Pipeline for Particle–Magnet Feature-Space Visualization

    Jürgen Döllner, Willy Scheibel, and Jose Carvajal

    Proceedings of the 31st International Conference on 3D Web Technology, 2026.

    DOI: 10.1145/3822516.3837796 ↗
  3. ISPRS Annals · 2026

    Exploring Multivariate Geospatial Data Through Interactive Feature-Space Mapping and Visualization

    Jürgen Döllner and Josafat-Mattias Burmeister

    ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XII-4/W1-2026, pp. 89–96.

    DOI: 10.5194/isprs-annals-XII-4-W1-2026-89-2026 ↗