What type of study is this?

This is a Experimental Study study.

October 5, 2025Open Access

Are Time Series Foundation Models Susceptible to Catastrophic Forgetting?

NKNouha KaraouliInstitut de Recherche en Informatique et Systèmes Aléatoires DCDenis CoquenetCentre National de la Recherche Scientifique ÉFÉlisa FromontMaritime Administration of Latvia

Key Points

Fine-tuning TSFMs on new datasets often leads to significant performance degradation on previously learned tasks.
Experiments utilized synthetic datasets with varying periodic structures to explore the stability-plasticity trade-off.
This study highlights the under-explored issue of catastrophic forgetting in the context of continual learning in machine learning.
The findings point to critical implications for the robustness of TSFMs when adapting to new forecasting tasks.

Abstract

Time Series Foundation Models (TSFMs) have shown promising zero-shot generalization across diverse forecasting tasks. However, their robustness to continual adaptation remains underexplored. In this work, we investigate the extent to which TSFMs suffer from catastrophic forgetting when fine-tuned sequentially on multiple datasets. Using synthetic datasets designed with varying degrees of periodic structure, we measure the trade-off between adaptation to new data and retention of prior knowledge. Our experiments reveal that, while fine-tuning improves performance on new tasks, it often causes significant degradation on previously learned ones, illustrating a fundamental stability-plasticity dilemma.

Ask AI

Helpful

Bookmark

View Full Paper