Reproduce “Evolutional Deep Neural Network” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yifan Du, Tamer A. Zaki
An Evolutional Deep Neural Network (EDNN) framework models spatial solutions with dynamically evolving parameters, enabling stable and accurate long-term predictions of partial differential equations. This approach precisely embeds physical constraints and successfully simulates chaotic turbulent flows, accurately capturing their long-time statistics.
Key insights
- Decoupling the spatial and temporal representation within the network explicitly enforces causality and significantly reduces network complexity for extended prediction horizons compared to global spatio-temporal models.
- Embedding linear physical constraints directly into the network architecture guarantees their exact satisfaction, eliminating the need for approximate penalty terms and their associated tuning challenges.
- Dynamically marching network parameters allows for indefinitely long time predictions, enabling the accurate capture of statistically stationary states and long-time statistics in chaotic and turbulent systems.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.