Contrastive Meta Learning for Dynamical Systems

Wang Zhang, Ziwen Ma, Subhro Das, Lam M. Nguyen, Luca Daniel

Keywords: dynamical system, meta learning, contrastive learning

Abstract: Recent advancements in deep learning have significantly impacted the study of dynamical systems. Traditional approaches predominantly rely on supervised learning paradigms, limiting their scope to large scale problems and adaptability to new systems. This paper introduces a novel meta learning framework tailored for dynamical system forecasting, hinging on the concept of mapping the observed trajectories to a system-specific embedding space which encapsulates the inter-system characteristics and enriches the feature set for downstream prediction tasks. Central to our framework is the use of contrastive learning for trajectory data coupled with a series of neural network architecture designs to extract the features as augmented embedding for modeling system behavior. We present the application of zero-shot meta-learning to dynamical systems, demonstrating a substantial enhancement in performance metrics compared to existing baseline models. A notable byproduct of our methodology is the improved interpretability of the embeddings, which now carries explicit physical significance. Our results not only set a new benchmark in the field but also pave the way for enhanced interpretability and deeper understanding of complex dynamical systems, potentially opens new directions for how we approach system analysis and prediction.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7237