DualWeaver: Synergistic Feature Weaving Surrogates for Multivariate Forecasting with Univariate Time Series Foundation Models

TL;DR

DualWeaver introduces a novel framework that adapts univariate time-series foundation models for multivariate forecasting by using symmetric surrogate series and a feature-fusion module, improving accuracy and robustness.

Contribution

It proposes a structurally symmetric surrogate-based approach to extend univariate TSFMs to multivariate forecasting, with a regularization for robustness.

Findings

Outperforms state-of-the-art multivariate forecasters in accuracy.

Demonstrates improved stability across diverse datasets.

Effective in capturing cross-variable dependencies.

Abstract

Time-series foundation models (TSFMs) have achieved strong univariate forecasting through large-scale pre-training, yet effectively extending this success to multivariate forecasting remains challenging. To address this, we propose DualWeaver, a novel framework that adapts univariate TSFMs (Uni-TSFMs) for multivariate forecasting by using a pair of learnable, structurally symmetric surrogate series. Generated by a shared auxiliary feature-fusion module that captures cross-variable dependencies, these surrogates are mapped to TSFM-compatible series via the forecasting objective. The symmetric structure enables parameter-free reconstruction of final predictions directly from the surrogates, without additional parametric decoding. A theoretically grounded regularization term is further introduced to enhance robustness against adaptation collapse. Extensive experiments on diverse real-world datasets show that DualWeaver outperforms state-of-the-art multivariate forecasters in both accuracy and stability. We release the code at https://github.com/li-jinpeng/DualWeaver.