WaveMoE: A Wavelet-Enhanced Mixture-of-Experts Foundation Model for Time Series Forecasting

TL;DR

WaveMoE introduces a wavelet-enhanced mixture-of-experts model that integrates frequency-domain information into time series forecasting, improving performance on diverse benchmarks.

Contribution

It presents a novel dual-path architecture combining time series and wavelet tokens with shared expert routing for scalable, frequency-aware forecasting.

Findings

Improves forecasting accuracy on 16 benchmark datasets.

Effectively models complex temporal patterns like periodicity.

Demonstrates potential of wavelet integration in foundation models.

Abstract

Time series foundation models (TSFMs) have recently achieved remarkable success in universal forecasting by leveraging large-scale pretraining on diverse time series data. Complementing this progress, incorporating frequency-domain information yields promising performance in enhancing the modeling of complex temporal patterns, such as periodicity and localized high-frequency dynamics, which are prevalent in real-world time series. To advance this direction, we propose a new perspective that integrates explicit frequency-domain representations into scalable foundation models, and introduce WaveMoE, a wavelet-enhanced mixture-of-experts foundation model for time series forecasting. WaveMoE adopts a dual-path architecture that jointly processes time series tokens and wavelet tokens aligned along a unified temporal axis, and coordinates them through a shared expert routing mechanism that enables consistent expert specialization while efficiently scaling model capacity. Preliminary experimental results on 16 diverse benchmark datasets indicate that WaveMoE has the potential to further improve forecasting performance by incorporating wavelet-domain corpora.