DLA-VPS: Deep-Learning-Assisted Visual Parameter Space Analysis of Cosmological Simulations

Area of Science:

• Cosmology
• Computational Astrophysics
• Data Science

Background:

• Cosmological simulations are crucial for understanding the universe but are computationally expensive.
• Analyzing large parameter spaces requires numerous simulation runs, hindering research.
• Efficient parameter space exploration is vital for extracting meaningful insights from cosmological data.

Purpose of the Study:

• To develop an interactive visual system for efficient understanding of cosmological parameter spaces.
• To reduce the computational cost associated with high-fidelity universe simulations.
• To facilitate the selection of optimal simulation input parameters for analysis tasks.

Main Methods:

• Utilized a Generative Adversarial Network (GAN)-based surrogate model to reconstruct simulation outputs.
• Employed deep neural network insights for enhanced parameter space exploration.
• Integrated an interactive visual system for user-guided analysis.

Main Results:

• The system successfully reconstructs simulation outputs without running expensive simulations.
• Information extracted from surrogate models aids in understanding complex parameter landscapes.
• Case studies demonstrate effective identification of valuable simulation parameters and subregions.

Conclusions:

• The proposed system significantly accelerates the analysis of cosmological data by reducing simulation time.
• Interactive visualization coupled with AI-driven surrogate models enhances parameter space exploration.
• This approach offers a powerful tool for cosmologists to efficiently analyze simulation results and discover key parameter configurations.