CoRA: Boosting Time Series Foundation Models for Multivariate Forecasting through Correlation-aware Adapter

2026-03-23

AI Summary

Existing foundation models for time series struggle to effectively capture the complex inter-variable correlations in multivariate forecasting, limiting their predictive performance. To address this limitation, this work proposes CoRA—a lightweight, plug-and-play, correlation-aware adapter that enhances a model’s ability to represent variable dependencies during fine-tuning. CoRA innovatively decomposes the correlation matrix into time-varying and time-invariant low-rank components and introduces learnable polynomials to model dynamic correlations. Furthermore, it incorporates a dual contrastive learning mechanism with a heterogeneous partial contrastive loss to discern positive and negative correlation patterns—all without introducing additional inference overhead. Extensive experiments on ten real-world datasets demonstrate that CoRA significantly improves the multivariate forecasting accuracy of state-of-the-art time series foundation models.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Multi-modal Vision
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Most existing Time Series Foundation Models (TSFMs) use channel independent modeling and focus on capturing and generalizing temporal dependencies, while neglecting the correlations among channels or overlooking the different aspects of correlations. However, these correlations play a vital role in Multivariate time series forecasting. To address this, we propose a CoRrelation-aware Adapter (CoRA), a lightweight plug-and-play method that requires only fine-tuning with TSFMs and is able to capture different types of correlations, so as to improve forecast performance. Specifically, to reduce complexity, we innovatively decompose the correlation matrix into low-rank Time-Varying and Time-Invariant components. For the Time-Varying component, we further design learnable polynomials to learn dynamic correlations by capturing trends or periodic patterns. To learn positive and negative correlations that appear only among some channels, we introduce a novel dual contrastive learning method that identifies correlations through projection layers, regulated by a Heterogeneous-Partial contrastive loss during training, without introducing additional complexity in the inference stage. Extensive experiments on 10 real-world datasets demonstrate that CoRA can improve TSFMs in multivariate forecasting performance.

Problem

• Multivariate Time Series Forecasting
• Time Series Foundation Models
• Channel Correlation
• Temporal Dependencies
• Correlation Modeling

Innovation

• Correlation-aware Adapter
• Multivariate Time Series Forecasting
• Low-rank Decomposition
• Dual Contrastive Learning
• Time Series Foundation Models