<ul><li>Time Series Foundation Models (TSFMs) struggle to outperform smaller, specialized models even after fine-tuning on specific data.</li><li>Empirical studies show that TSFMs have inherent sparsity and redundancy, indicating adaptability to diverse tasks.</li><li>A structured pruning method is proposed to enhance adaptation of TSFMs by focusing on relevant network substructures during fine-tuning.</li><li>Experiments on seven TSFMs and six benchmarks reveal that pruning and then fine-tuning these models leads to improved forecasting performance, surpassing specialized baselines.</li></ul>

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

Discover more