Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Learning to Emulate Chaos: Adversarial Optimal Transport Regularization

About

Chaos arises in many complex dynamical systems, from weather to power grids, but is difficult to accurately model using data-driven emulators, including neural operator architectures. For chaotic systems, the inherent sensitivity to initial conditions makes exact long-term forecasts theoretically infeasible, meaning that traditional squared-error losses often fail when trained on noisy data. Recent work has focused on training emulators to match the statistical properties of chaotic attractors by introducing regularization based on handcrafted local features and summary statistics, as well as learned statistics extracted from a diverse dataset of trajectories. In this work, we propose a family of adversarial optimal transport objectives that jointly learn high-quality summary statistics and a physically consistent emulator. We theoretically analyze and experimentally validate a Sinkhorn divergence formulation (2-Wasserstein) and a WGAN-style dual formulation (1-Wasserstein). Our experiments across a variety of chaotic systems, including systems with high-dimensional chaotic attractors, show that emulators trained with our approach exhibit significantly improved long-term statistical fidelity.

Gabriel Melo, Leonardo Santiago, Peter Y. Lu• 2026

Related benchmarks

TaskDatasetResultRank
Statistical Emulation of Chaotic SystemsKS single-traj (Noisy (sigma = 0.3))
L1 Histogram Distance0.259
9
Statistical Emulation of Chaotic SystemsL96 single-traj Clean (sigma = 0.0)
L1 Histogram Distance0.082
9
Statistical Emulation of Chaotic SystemsL96 multi-traj Noisy (sigma = 0.3)
L1 Histogram Distance0.149
5
Statistical Emulation of Chaotic SystemsL96 multi-traj (Clean (sigma = 0.0))
L1 Histogram Distance0.083
5
Leading Lyapunov Exponent EstimationLorenz-96 (L96) single-trajectory
LLE2.336
4
Statistical Emulation of Chaotic SystemsL96 single-traj (Noisy (sigma = 0.3))
L1 Histogram Distance0.151
4
Statistical Emulation of Chaotic SystemsKolmogorov Flow single-traj Clean (sigma = 0.0)
L1 Histogram Distance0.178
4
Showing 7 of 7 rows

Other info

Follow for update