Learning to Emulate Chaos: Adversarial Optimal Transport Regularization
Researchers have developed a new method using adversarial optimal transport to improve the accuracy of data-driven emulators for chaotic systems. Traditional methods struggle with the inherent sensitivity of chaotic systems, leading to poor long-term forecasts. This novel approach learns better summary statistics and creates physically consistent emulators, showing improved statistical fidelity across various chaotic systems, including high-dimensional ones. AI
IMPACT Introduces a novel regularization technique for improving emulator fidelity in chaotic systems, potentially impacting scientific modeling.