Research map / Time series / Causal and generative time series
Uncertainty-DTW for Time Series and Sequences
Lei Wang, Piotr Koniusz
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 SigmaNet-predicted uncertainty instead of a free-variable Sigma (ablation) | uDTW · NTU-60 supervised (50 classes) | accuracy: 68.0 -> 72.4 | Improved |
| add SigmaNet-predicted uncertainty instead of a free-variable Sigma (ablation) | uDTW · NTU-60 supervised (10 classes) | accuracy: 54.1 -> 56.9 | Improved |
| add SigmaNet-predicted uncertainty instead of a free-variable Sigma (ablation) | uDTW · UCR nearest neighbor K=1 | accuracy: 77.0 -> 80.0 | Improved |
| add SigmaNet-predicted uncertainty instead of a free-variable Sigma (ablation) | uDTW · UCR nearest centroid | accuracy: 70.9 -> 72.2 | Improved |
| add aleatoric uncertainty weighting to soft-DTW divergence | sDTW div. · UCR nearest neighbor K=5 | accuracy: 80.1 -> 83.3 | Improved |
| add aleatoric uncertainty weighting to soft-DTW divergence | sDTW div. · UCR nearest centroid | accuracy: 70.9 -> 72.2 | Improved |
| train forecaster with uDTW loss instead of sDTW div. | MLP forecaster · ECG5000 | MSE: 24.6 -> 23.0 | Improved |
| train forecaster with uDTW loss instead of Euclidean | MLP forecaster · ECG5000 | MSE: 32.1 -> 23.0 | Improved |
| add uDTW in place of sDTW div. for few-shot recognition | few-shot AR · NTU-60 supervised (50 classes) | accuracy: 69.0 -> 72.4 | Improved |
| add uDTW in place of TAP | few-shot AR · NTU-120 supervised (100 classes) | accuracy: 47.3 -> 49.0 | Improved |
| add uDTW in place of sDTW div. in unsupervised few-shot | few-shot AR · Kinetics-skeleton 3D unsupervised | accuracy: 28.9 -> 32.7 | Improved |
| derive uDTW from Laplace instead of Normal likelihood (ablation) | uDTW · NTU-60 supervised (50 classes) | accuracy: 72.4 -> 70.3 | Hurt |
Closest papers by meaning
- Perturbed Dynamic Time Warping: A Probabilistic Framework and Generalized Variants (ICLR 2026 · similarity 0.88)
- Deep Declarative Dynamic Time Warping for End-to-End Learning of Alignment Paths (ICLR 2023 · similarity 0.87)
- Soft-DTW: a Differentiable Loss Function for Time-Series (ICML 2017 · similarity 0.86)
- Fortifying Time Series: DTW-Certified Robust Anomaly Detection (NeurIPS 2025 · similarity 0.85)
- Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration (AAAI 2024 · similarity 0.85)
- Improved Learning via k-DTW: A Novel Dissimilarity Measure for Curves (ICML 2025 · 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.
