Research map / Architectures and efficiency / Pruning and efficient tuning
Lost in the Non-convex Loss Landscape: How to Fine-tune the Large Time Series Model?
Xu Zhang, Peng Wang, Wei Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace full fine-tuning with smoothed full fine-tuning (SFF) | Timer · TSF datasets (horizon 96) | MSE: average 3% and up to 6.5% MSE reduction | Improved |
| replace full fine-tuning with SFF | Timer · ETTh2 (100% data) | MSE: 0.3047 -> 0.2737 | Improved |
| replace full fine-tuning with SFF | TimesFM · TSF datasets | MSE: average improvement of 11.45% | Improved |
| replace full fine-tuning with SFF | MOMENT · TSF datasets | MSE: average improvement of 8.31% | Improved |
| replace full fine-tuning with SFF | Timer · Anomaly detection Group 1 | predicted MSE on anomalous segments: 0.072 -> 0.136 | Improved |
| replace linear probing then fine-tuning with SFF | Timer · TSF Exchange (avg 1-3% data) | MSE: 0.4801 -> 0.0856 | Improved |
| add weight smoothing before zero-shot inference | Timer · zero-shot TSF, seven datasets | MSE: average gain of 6.13% | Improved |
| add weight smoothing before zero-shot inference | TimesFM · zero-shot TSF, seven datasets | MSE: average gain of 35.75% | Improved |
| set interpolation coefficient alpha near 0.85 (ablation) | SFF · zero-shot TSF horizon 96 | MSE: alpha of 0.85 gives best zero-shot performance | Improved |
| add smoothing to Timer fine-tuning (ablation) | Timer · Test MSE convergence | convergence epochs: both converge within the first epoch | No clear effect |
| replace full fine-tuning with SFF | TTMs · ETTm1 (horizon 96) | MSE: 0.308 -> 0.309 | No clear effect |
| replace full fine-tuning with SFF | UniTS · Electricity (horizon 96) | MSE: 0.476 -> 0.309 | Improved |
Closest papers by meaning
- Multi-Scale Finetuning for Encoder-based Time Series Foundation Models (NeurIPS 2025 · similarity 0.89)
- Less is More: Unlocking Specialization of Time Series Foundation Models via Structured Pruning (NeurIPS 2025 · similarity 0.88)
- In-Context Fine-Tuning for Time-Series Foundation Models (ICML 2025 · similarity 0.88)
- Timer: Generative Pre-trained Transformers Are Large Time Series Models (ICML 2024 · similarity 0.88)
- Large Pre-trained time series models for cross-domain Time series analysis tasks (NeurIPS 2024 · similarity 0.87)
- Linearization Explains Fine-Tuning in Large Language Models (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
