Research map / Time series / Foundation models for time series
TSPulse: Tiny Pre-Trained Models with Disentangled Representations for Rapid Time-Series Analysis
Vijay Ekambaram, Subodh Kumar, Arindam Jati, Sumanta Mukherjee, Tomoya Sakai, Pankaj Dayama, Wesley M. Gifford, Jayant Kalagnanam
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 |
|---|---|---|---|
| use multi-head triangulation instead of ensemble head (ablation) | TSPulse zero-shot AD · TSB-AD-U | VUS-PR: 0.44 -> 0.48 | Improved |
| use multi-head triangulation instead of ensemble head (ablation) | TSPulse zero-shot AD · TSB-AD-M | VUS-PR: 0.31 -> 0.36 | Improved |
| add TSLens instead of average pooling (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.675 -> 0.747 | Improved |
| add TSLens instead of max pooling (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.645 -> 0.747 | Improved |
| keep short embedding (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.689 -> 0.747 | Improved |
| keep long embedding (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.681 -> 0.747 | Improved |
| use masking during fine-tuning (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.691 -> 0.747 | Improved |
| identity init of channel-mixing blocks instead of random (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.685 -> 0.747 | Improved |
| add dual-space (time+FFT) learning (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.696 -> 0.747 | Improved |
| add virtual channel expansion (ablation) | TSPulse classification · UEA (17 datasets) | accuracy: 0.734 -> 0.747 | Improved |
| add dual-space learning (ablation) | TSPulse zero-shot imputation · hybrid-mask imputation | MSE: 0.081 -> 0.074 | Improved |
| pre-train with hybrid masking instead of block-only masking (ablation) | TSPulse zero-shot imputation · hybrid-mask imputation | MSE: 0.354 -> 0.074 | Improved |
Closest papers by meaning
- Exploring Representations and Interventions in Time Series Foundation Models (ICML 2025 · similarity 0.87)
- TSLANet: Rethinking Transformers for Time Series Representation Learning (ICML 2024 · similarity 0.87)
- T-Rep: Representation Learning for Time Series using Time-Embeddings (ICLR 2024 · similarity 0.86)
- TimeSiam: A Pre-Training Framework for Siamese Time-Series Modeling (ICML 2024 · similarity 0.85)
- Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning (AAAI 2025 · similarity 0.85)
- $\alpha$TC-VAE: On the relationship between Disentanglement and Diversity (ICLR 2024 · similarity 0.85)
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.
