Publications

Scale-Space Splatting: Reforming Spacetime for Cross-Scale Exploration of Integral Measures in Molecular Dynamics

J. Pálenik, J. Byška, S. Bruckner, and H. Hauser

Abstract

Understanding large amounts of spatiotemporal data from particle-based simulations, such as molecular dynamics, often relies on the computation and analysis of aggregate measures. These, however, by virtue of aggregation, hide structural information about the space/time localization of the studied phenomena. This leads to degenerate cases where the measures fail to capture distinct behaviour. In order to drill into these aggregate values, we propose a multi-scale visual exploration technique. Our novel representation, based on partial domain aggregation, enables the construction of a continuous scale-space for discrete datasets and the simultaneous exploration of scales in both space and time. We link these two scale-spaces in a scale-space space-time cube and model linked views as orthogonal slices through this cube, thus enabling the rapid identification of spatio-temporal patterns at multiple scales. To demonstrate the effectiveness of our approach, we showcase an advanced exploration of a protein-ligand simulation.

J. Pálenik, J. Byška, S. Bruckner, and H. Hauser, "Scale-Space Splatting: Reforming Spacetime for Cross-Scale Exploration of Integral Measures in Molecular Dynamics," IEEE Transactions on Visualization and Computer Graphics, vol. 26, iss. 1, p. 643–653, 2020. doi:10.1109/TVCG.2019.2934258
[BibTeX]

Understanding large amounts of spatiotemporal data from particle-based simulations, such as molecular dynamics, often relies on the computation and analysis of aggregate measures. These, however, by virtue of aggregation, hide structural information about the space/time localization of the studied phenomena. This leads to degenerate cases where the measures fail to capture distinct behaviour. In order to drill into these aggregate values, we propose a multi-scale visual exploration technique. Our novel representation, based on partial domain aggregation, enables the construction of a continuous scale-space for discrete datasets and the simultaneous exploration of scales in both space and time. We link these two scale-spaces in a scale-space space-time cube and model linked views as orthogonal slices through this cube, thus enabling the rapid identification of spatio-temporal patterns at multiple scales. To demonstrate the effectiveness of our approach, we showcase an advanced exploration of a protein-ligand simulation.
@ARTICLE{Palenik-2019-Splatting,
author={J. P\'{a}lenik and J. By\v{s}ka and S. Bruckner and H. Hauser},
journal={IEEE Transactions on Visualization and Computer Graphics},
title={Scale-Space Splatting: Reforming Spacetime for Cross-Scale Exploration of Integral Measures in Molecular Dynamics},
year={2020},
volume={26},
number={1},
pages={643--653},
keywords={Data visualization;Computational modeling;Time series analysis;Atmospheric measurements;Particle measurements;Analytical models;Kernel;Scale space;time-series;scientific simulation;multi-scale analysis;space-time cube;molecular dynamics},
doi={10.1109/TVCG.2019.2934258},
ISSN={1077-2626},
month={},
pdf = "pdfs/scale-space-splatting.pdf",
images = "images/scale-space-teaser.png",
thumbnails = "images/scale-space-teaser-thumb.png",
abstract = "Understanding large amounts of spatiotemporal data from particle-based simulations, such as molecular dynamics, often relies on the computation and analysis of aggregate measures. These, however, by virtue of aggregation, hide structural information about the space/time localization of the studied phenomena. This leads to degenerate cases where the measures fail to capture distinct behaviour. In order to drill into these aggregate values, we propose a multi-scale visual exploration technique. Our novel representation, based on partial domain aggregation, enables the construction of a continuous scale-space for discrete datasets and the simultaneous exploration of scales in both space and time. We link these two scale-spaces in a scale-space space-time cube and model linked views as orthogonal slices through this cube, thus enabling the rapid identification of spatio-temporal patterns at multiple scales. To demonstrate the effectiveness of our approach, we showcase an advanced exploration of a protein-ligand simulation.",
}
projectidprojectid

Media

Downloads

Full paper [PDF]