Associate Professor, Shanghai Jiao Tong University
2 papers at NeurIPS 2025
We propose a prune-then-finetune paradigm to specialize time series foundation models in both levels of architecture and parameter weights, effectively improving downstream forecasting performance.