Probabilistic Data-Driven Sampling via Multi-Criteria Importance Analysis.

Area of Science:

• Computational Science
• Data Science
• High-Performance Computing

Background:

• Supercomputer capabilities are rapidly advancing, enabling highly detailed simulations.
• Input/Output (I/O) performance has not kept pace, creating a data storage bottleneck.
• Traditional data saving methods are insufficient for novel simulation features.

Purpose of the Study:

• To develop a novel data sampling scheme for high-performance computing.
• To address the challenge of storing massive simulation datasets.
• To enable the capture of important, previously unseen simulation features.

Main Methods:

• Proposing a data-driven intelligent sampling strategy.
• Selecting data points based on unusual values and high gradients.
• Reducing data size by orders of magnitude while preserving critical information.

Main Results:

• The novel sampling scheme significantly reduces data volume.
• Important simulation regions are effectively preserved.
• The proposed method outperforms traditional sampling techniques.

Conclusions:

• Intelligent data sampling is crucial for managing large-scale simulations.
• The developed approach offers an effective solution for data reduction.
• This method enhances the ability to analyze complex simulation outputs.