A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems [R]

A new NeurIPS 2026 paper, "A Minimal Interpretable Architecture for Zero-Shot Reconstruction of Dynamical Systems," introduces a simplified approach to dynamical systems (DS) foundation models. The research focuses on identifying the minimal necessary components to accurately replicate the long-term statistical and geometrical characteristics of DS. This work, detailed in a preprint available on arXiv (2607.14937), aims to enhance the interpretability and efficiency of DS reconstruction.

This paper uniquely focuses on reducing a dynamical systems foundation model to its minimally necessary ingredients, unlike other approaches that may add complexity.