Research map / Time series / Foundation models for time series
Unsupervised Feature Extraction by Time-Contrastive Learning and Nonlinear ICA
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Adding TCL-trained nonlinear feature extractor followed by linear ICA (vs. baseline methods) on simulated nonlinear mixtures | nonlinear ICA with nonstationary Laplacian sources · simulation, L>1 nonlinear mixing | mean absolute correlation with true q(s) | Helped |
| Using TCL instead of NSVICA (linear ICA on nonstationary variance) in the linear mixing case | simulation, L=1 · simulation, linear mixing | mean absolute correlation with true q(s) | Hurt |
| Increasing the number of layers in the mixing MLP (stronger nonlinearity) for TCL (ablation) | TCL feature extractor · simulation | classification accuracy and source recovery correlation | Hurt |
| Increasing the number of segments (more data) for TCL (ablation) | TCL · simulation | mean absolute correlation with true q(s) | Helped |
| Using multi-task MLR with separate top-layer classifier per session and subject (shared feature-MLP) instead of one MLR over all segments (ablation) | TCL on MEG · MEG resting-state training, task-session classification | classification accuracy | Helped |
| Adding hidden layers (multi-layer MLP, L>=3 vs linear L=1) to TCL feature extractor (ablation) | TCL on MEG · MEG task-session modality classification | classification accuracy | Helped |
| TCL features versus baseline methods (log-power features of other methods) on MEG | TCL feature extractor · MEG task-session modality classification, 4 classes | classification accuracy | Helped |
Closest papers by meaning
- On the Identifiability of Nonlinear ICA: Sparsity and Beyond (NeurIPS 2022 · similarity 0.87)
- Robustness of Nonlinear Representation Learning (ICML 2024 · similarity 0.84)
- Feature learning from non-Gaussian inputs: the case of Independent Component Analysis in high dimensions (ICML 2025 · similarity 0.84)
- TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning (AAAI 2024 · similarity 0.83)
- Cross-Domain Contrastive Learning for Time Series Clustering (AAAI 2024 · similarity 0.82)
- Parametric Augmentation for Time Series Contrastive Learning (ICLR 2024 · similarity 0.82)
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.