Research map / Time series / Foundation models for time series
Scale-teaching: Robust Multi-scale Training for Time Series Classification with Noisy Labels
Zhen Liu, Peitian Ma, Dongliang Chen, Wenbin Pei, Qianli Ma
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 |
|---|---|---|---|
| remove cross-scale fusion (ablation) | Scale-teaching · UniMiB-SHAR Sym 50% | accuracy (%): 73.32 -> 81.31 | Improved |
| remove cross-scale fusion (ablation) | Scale-teaching · HAR Sym 50% | accuracy (%): 88.47 -> 90.17 | Improved |
| use only a single scale instead of multi-scale (ablation) | Scale-teaching · UniMiB-SHAR Asym 40% | accuracy (%): 60.32 -> 70.68 | Improved |
| add graph-learning label correction (ablation) | Scale-teaching · HAR Sym 50% | accuracy (%): 88.06 -> 90.17 | Improved |
| add momentum embedding stabilization (ablation) | Scale-teaching · UniMiB-SHAR Sym 50% | accuracy (%): 80.57 -> 81.31 | Improved |
| add dynamic threshold for pseudo-label selection (ablation) | Scale-teaching · UniMiB-SHAR Sym 50% | accuracy (%): 77.42 -> 81.31 | Improved |
| add cross-scale fusion to coarse scale classifier (ablation) | Scale-teaching · UCR 128 Sym 20% | avg accuracy (%): 28.17 -> 68.70 | Improved |
| add cross-scale fusion to fine scale classifier (ablation) | Scale-teaching · UCR 128 Sym 20% | avg accuracy (%): 65.13 -> 59.67 | Hurt |
| Scale-teaching vs CULCU under Asym 40% noise | Scale-teaching · HAR | accuracy (%): 87.22 -> 89.62 | Improved |
| Scale-teaching vs CULCU under Sym 50% noise | Scale-teaching · HAR | accuracy (%): 89.91 -> 90.17 | No clear effect |
| Scale-teaching vs Standard training under Sym 20% noise | Scale-teaching · UniMiB-SHAR | accuracy (%): 87.07 -> 90.69 | Improved |
| Scale-teaching vs SREA/Mixup without noise | Scale-teaching · HAR (clean labels) | accuracy (%): 95.42 -> 94.72 | No clear effect |
Closest papers by meaning
- SVGL: Scale-Variable Graph Learning in Model Space for Multivariate Time Series Classification (AAAI 2026 · similarity 0.87)
- Training Robust Deep Models for Time-Series Domain: Novel Algorithms and Theoretical Analysis (AAAI 2022 · similarity 0.86)
- MPTSNet: Integrating Multiscale Periodic Local Patterns and Global Dependencies for Multivariate Time Series Classification (AAAI 2025 · similarity 0.85)
- Learning under Temporal Label Noise (ICLR 2025 · similarity 0.85)
- Hi-Patch: Hierarchical Patch GNN for Irregular Multivariate Time Series (ICML 2025 · similarity 0.84)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.84)
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.
