Immersive Visual Cues for Understanding of Nonlinear Dimensionality Reduction in Mobile VR

Résumé fourni par la source

Nonlinear dimensionality reduction (NLDR) is widely used to project high-dimensional data into low-dimensional embeddings for visualization and analysis. However, these embeddings are often difficult to interpret, particularly for non-expert users. Existing studies have mainly examined how dimensionality reduction results can be interpreted through conventional visualizations, while supporting NLDR understanding through immersive visualizations has received far less attention. In this paper, we present an immersive visualization system that features three immersive visual cues for interpreting NLDR embeddings: a feature attribution panel, local reliability highlighting, and neighbor relationship visualization. The system was implemented as a standalone application on a mobile VR device and evaluated through a within-subject user study with 19 participants. Based on their task performance and subjective questionnaire responses, the results show that the proposed visual cues support selected NLDR analysis tasks and improve perceived confidence in solving tasks, although the effects were not equally strong across all tasks. Overall, this work demonstrates the potential of mobile VR as a practical setting for human-centered, interpretation-oriented immersive analytics of NLDR results.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.