Research map / Time series / Foundation models for time series
Regularization-free Diffeomorphic Temporal Alignment Nets
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 |
|---|---|---|---|
| Replace WCSS+smoothness regularization with regularization-free ICAE loss | DTAN_DIFW with single HP config · UCR 84 datasets | NCC median accuracy: 0.604 -> 0.665 | Helped |
| Add centroid triplet loss on top of ICAE (ablation) | DTAN_DIFW LICAE · UCR 84 datasets | NCC median accuracy: 0.665 -> 0.707 | Helped |
| Replace WCSS+reg with ICAE loss | DTAN_DIFW single HP config · UCR 117 datasets | NCC median accuracy: 0.603 -> 0.656 | Helped |
| Remove regularization from WCSS loss (ablation) | DTAN_DIFW WCSS · UCR 117 datasets | NCC median accuracy: 0.603 -> 0.609 | Helped |
| Add centroid triplet loss to ICAE (full 128 datasets) (ablation) | DTAN_DIFW LICAE · UCR 128 datasets | NCC median accuracy: 0.623 -> 0.67 | Helped |
| Switch backbone from TCN to InceptionTime with ICAE-triplet (ablation) | DTAN LICAE-triplet · UCR 127 datasets | NCC: 0.632 -> 0.67 | Helped |
| Add smoothness regularization to WCSS-triplet (ablation) | DTAN_DIFW WCSS-triplet · UCR 117 datasets | NCC: 0.642 -> 0.603 | Hurt |
| Regularization-free ICAE vs tuned-HP DTAN (full search) | DTAN_DIFW · UCR 84 datasets, HP search allowed | NCC median accuracy: 0.749 (WCSS+REG, 96 configs) vs 0.665 (LICAE, single config) | Hurt |
| Warp-based ICAE loss applied to multivariate SpokenArabicDigits | DTAN with LICAE · SpokenArabicDigits | NCC accuracy: 0.08 -> 0.402 | Helped |
Closest papers by meaning
- Diffeomorphic Temporal Alignment Nets (NeurIPS 2019 · similarity 0.94)
- Closed-Form Diffeomorphic Transformations for Time Series Alignment (ICML 2022 · similarity 0.89)
- Deep Canonical Time Warping (CVPR 2016 · similarity 0.88)
- TimePoint: Accelerated Time Series Alignment via Self-Supervised Keypoint and Descriptor Learning (ICML 2025 · similarity 0.86)
- Deep Declarative Dynamic Time Warping for End-to-End Learning of Alignment Paths (ICLR 2023 · similarity 0.86)
- TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting (NeurIPS 2026 · similarity 0.83)
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.