Saved in:
Bibliographic Details
Main Authors: Zhu, Jiawen, Liu, Shuhan, Weng, Di, Wu, Yingcai
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.20678
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910239748521984
author Zhu, Jiawen
Liu, Shuhan
Weng, Di
Wu, Yingcai
author_facet Zhu, Jiawen
Liu, Shuhan
Weng, Di
Wu, Yingcai
contents Non-stationary time series forecasting is challenged by evolving distribution shifts that static models struggle to capture. While Mixture-of-Experts (MoE) architectures offer a promising paradigm for decoupling complex drift patterns, existing approaches are limited by fixed expert pools and memoryless routing, hampering their ability to adapt to abrupt regime shifts. To address this, we propose Dynamic TMoE, a framework that unifies architectural evolution with temporal continuity during learning phase. By detecting distribution shifts via Maximum Mean Discrepancy (MMD), we dynamically instantiate heterogeneous experts and prune redundant ones to optimize capacity. Additionally, a temporal memory router leverages recurrent states and an anomaly repository to ensure stable, context-aware expert selection without requiring test-time updates. Experiments on nine benchmarks demonstrate state-of-the-art performance, reducing MSE by 10.4% and MAE by 7.8%. Code is available at https://github.com/andone-07/Dynamic-TMoE.
format Preprint
id arxiv_https___arxiv_org_abs_2605_20678
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting
Zhu, Jiawen
Liu, Shuhan
Weng, Di
Wu, Yingcai
Machine Learning
Artificial Intelligence
Non-stationary time series forecasting is challenged by evolving distribution shifts that static models struggle to capture. While Mixture-of-Experts (MoE) architectures offer a promising paradigm for decoupling complex drift patterns, existing approaches are limited by fixed expert pools and memoryless routing, hampering their ability to adapt to abrupt regime shifts. To address this, we propose Dynamic TMoE, a framework that unifies architectural evolution with temporal continuity during learning phase. By detecting distribution shifts via Maximum Mean Discrepancy (MMD), we dynamically instantiate heterogeneous experts and prune redundant ones to optimize capacity. Additionally, a temporal memory router leverages recurrent states and an anomaly repository to ensure stable, context-aware expert selection without requiring test-time updates. Experiments on nine benchmarks demonstrate state-of-the-art performance, reducing MSE by 10.4% and MAE by 7.8%. Code is available at https://github.com/andone-07/Dynamic-TMoE.
title Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2605.20678