Integrated Multi-system Prediction via Equilibrium State Evaluation

Beinan Xu, Andy Song, Jiti Gao

Keywords: Multi-system, Equilibrium, Prediction

TL;DR: A new method that can predict for multiple systems in just one run.

Abstract: This study presents a new paradigm of prediction, Equilibrium State Evaluation (ESE), which excels in multi-system prediction where systems interact with each other and every system needs its own prediction. Unlike mainstream prediction approaches, ESE views each system as an integral part under one structure and predicts all systems simultaneously in one go. It evaluates these systems' equilibrium state by analyzing the dynamics of their attributes in a holistic manner, instead of treating each system as an individual time series. The effectiveness of ESE is verified in synthetic and real world scenarios, in particular COVID-19 transmission, where each geographic region can be viewed as a system. So cases spreading across regions against the medical competency and demographic traits of these regions can be considered as an equilibrium problem rather than a time series problem. Extensive analysis and experiments show that ESE is linear in complexity and can be 10+ times faster than SOTA methods, yet achieving comparable or better prediction accuracy. More importantly, ESE can be integrated with these prediction methods to achieve both high accuracy and high speed, making it a powerful prediction mechanism, especially for scenarios that involve multiple systems. When the dimensionality of the multi-system increases, e.g. more systems joining, the advantages of ESE would become even more apparent.

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: 12376