Research map / Time series / Foundation models for time series
Self-Supervised Contrastive Representation Learning for Semi-Supervised Time-Series Classification
Emadeldeen Eldele, Mohamed Ragab, Zhenghua Chen, Min Wu, Chee Keong Kwoh, Xiaoli Li, Cuntai Guan
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 |
|---|---|---|---|
| add cross-view prediction (strong context predicts weak future and vice versa) (ablation) | TS-TCC temporal contrasting · HAR (linear eval, 5% labels) | accuracy: 68.16 -> 74.22 | Improved |
| add contextual contrasting module (ablation) | TS-TCC · HAR (linear eval, 5% labels) | accuracy: 74.22 -> 77.58 | Improved |
| add contextual contrasting module (ablation) | TS-TCC · Sleep-EDF (linear eval, 5% labels) | MF1-score: 61.28 -> 70.94 | Improved |
| replace unsupervised contextual contrasting with supervised contrastive loss on pseudo labels (ablation) | CA-TCC · HAR (linear eval, 5% labels) | accuracy: 77.58 -> 88.27 | Improved |
| use weak plus strong augmentation instead of two weak views (ablation) | TS-TCC · HAR (linear eval, 5% labels) | accuracy: 67.39 -> 77.58 | Improved |
| use weak plus strong augmentation instead of two strong views (ablation) | TS-TCC · HAR (linear eval, 5% labels) | accuracy: 50.37 -> 77.58 | Improved |
| use weak plus strong augmentation instead of one family on both views (ablation) | TS-TCC · Epilepsy (linear eval, 5% labels) | accuracy: 93.22 -> 93.12 | No clear effect |
| add jitter to scaling as the weak augmentation (ablation) | TS-TCC · HAR (linear eval, 5% labels, no strong aug) | accuracy: 46.12 -> 56.98 | Improved |
| add jitter to permutation as the strong augmentation (ablation) | TS-TCC · HAR (linear eval, 5% labels, no weak aug) | accuracy: 62.07 -> 72.35 | Improved |
| add jitter to time shift as the weak augmentation (ablation) | TS-TCC · HAR (linear eval, 5% labels) | accuracy: 71.57 -> 74.69 | Improved |
| use TS-TCC instead of SimCLR for unsupervised pretraining in Phase 1 (ablation) | CA-TCC · HAR (semi-supervised, 1% labels) | accuracy: 73.2 -> 77.3 | Improved |
| use class-aware TS-TCC instead of class-aware SimCLR in Phase 4 (ablation) | CA-TCC with TS-TCC pretraining · HAR (semi-supervised, 1% labels) | accuracy: 75.0 -> 77.3 | Improved |
Closest papers by meaning
- TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning (AAAI 2024 · similarity 0.89)
- Graph-Aware Contrasting for Multivariate Time-Series Classification (AAAI 2024 · similarity 0.88)
- Parametric Augmentation for Time Series Contrastive Learning (ICLR 2024 · similarity 0.88)
- Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning (AAAI 2025 · similarity 0.88)
- Soft Contrastive Learning for Time Series (ICLR 2024 · similarity 0.88)
- T-Rep: Representation Learning for Time Series using Time-Embeddings (ICLR 2024 · similarity 0.88)
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.
