<ul><li>arXiv:2505.23863v1 announces a new framework called PhyxMamba that integrates a Mamba-based state-space model with physics-informed principles for long-term forecasting of chaotic systems.</li><li>PhyxMamba reconstructs the attractor manifold from brief observations using time-delay embeddings to extract global dynamical features essential for accurate forecasting of chaotic systems.</li><li>The framework utilizes a generative training scheme that enables Mamba to replicate the physical process while incorporating multi-token prediction and attractor geometry regularization for enhanced prediction accuracy and preservation of key statistical invariants.</li><li>Extensive evaluations on various simulated and real-world chaotic systems demonstrate that PhyxMamba provides superior long-term forecasting capabilities and faithfully captures essential dynamical invariants from short-term data, applicable to fields such as climate science, neuroscience, and epidemiology.</li></ul>

Mamba Integrated with Physics Principles Masters Long-term Chaotic System Forecasting

Discover more