Low-Rank Adaptation of Time Series Foundational Models for Out-of-Domain Modality Forecasting

TL;DR

This paper explores applying Low-Rank Adaptation (LoRA) to time series foundational models, demonstrating improved out-of-domain forecasting performance in ICU vital signs with reduced fine-tuning costs.

Contribution

It is the first to systematically evaluate LoRA's effectiveness on time series foundational models like Lag-Llama, MOIRAI, and Chronos for out-of-domain forecasting tasks.

Findings

LoRA fine-tuning significantly improves forecasting accuracy.

Models achieve performance comparable to state-of-the-art trained from scratch.

Ablation studies reveal trade-offs between parameter count and performance.

Abstract

Low-Rank Adaptation (LoRA) is a widely used technique for fine-tuning large pre-trained or foundational models across different modalities and tasks. However, its application to time series data, particularly within foundational models, remains underexplored. This paper examines the impact of LoRA on contemporary time series foundational models: Lag-Llama, MOIRAI, and Chronos. We demonstrate LoRA's fine-tuning potential for forecasting the vital signs of sepsis patients in intensive care units (ICUs), emphasizing the models' adaptability to previously unseen, out-of-domain modalities. Integrating LoRA aims to enhance forecasting performance while reducing inefficiencies associated with fine-tuning large models on limited domain-specific data. Our experiments show that LoRA fine-tuning of time series foundational models significantly improves forecasting, achieving results comparable to state-of-the-art models trained from scratch on similar modalities. We conduct comprehensive ablation studies to demonstrate the trade-offs between the number of tunable parameters and forecasting performance and assess the impact of varying LoRA matrix ranks on model performance.