Gaussian Cubes: Real-Time Modeling for Visual Exploration of Large Multidimensional Datasets

Area of Science:

• Data Science
• Computer Science
• Statistics

Background:

• Current interactive systems for large datasets are limited to basic plots like histograms and heatmaps.
• Existing techniques struggle with complex analytical tasks on massive datasets in real time.

Purpose of the Study:

• To introduce Gaussian Cubes, a novel system for interactive modeling of very large datasets.
• To significantly enhance the analytical capabilities beyond current state-of-the-art visualization tools.

Main Methods:

• Gaussian Cubes precomputes multivariate Gaussian models for data subsets instead of counts.
• This approach enables interactive fitting of models like linear least squares and principal components analysis (PCA).

Main Results:

• Gaussian Cubes can fit hundreds of models on millions of data points in under a second.
• Demonstrated capabilities through case studies in earthquake simulations, astronomical catalogs, and transportation statistics.

Conclusions:

• Gaussian Cubes offers a powerful new paradigm for visual exploration and analysis of massive datasets.
• The system facilitates novel interactive modeling, significantly advancing the field of large-scale data analysis.