Generalized Prompt Tuning: How to Use a Frozen Pre-Trained Univariate Time Series Foundation Model for Multivariate Time Series Prediction

Time series foundation models are pre-trained on large datasets and are able to achieve state-of-the-art performance in diverse tasks. We observe that the majority of foundation models perform channel independence, where cross-channel correlations are ignored. In this study, we propose a prompt-tuning-inspired technique to perform channel-mixing for existing univariate time series pre-trained models, and compare the performance of different fine-tuning methods on time series foundation models.