Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning

Research on Time Series Foundation Models (TSFMs) shows that while they achieve strong zero-shot forecasting, fine-tuning often doesn't surpass smaller, specialized models. A proposed \"prune-then-finetune\" approach, which involves structured pruning before fine-tuning, significantly enhances TSFM performance, enabling state-of-the-art results across multiple benchmarks.