Research map / Time series / Long-term forecasting
Dynamic Semantic Tokenization for Time Series via Elastic Sampling on Physics-aware Perception
Huaizhang Liao, Zhixiong Yang, Jingyuan Xia, Yuheng Sun, Yue Zhang, Shengxi Li, Yongxiang Liu
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 physics-aware sampling (PA) to the tokenizer (ablation) | PATK · Epilepsy | accuracy: 79.13 -> 85.35 | Improved |
| add length adaptation re-weighting (LA) (ablation) | PATK · Epilepsy | accuracy: 85.35 -> 91.27 | Improved |
| add token significance re-weighting (TS) (ablation) | PATK · Epilepsy | accuracy: 91.27 -> 95.25 | Improved |
| add physics-aware loss (PAL) (ablation) | PATK · Epilepsy | accuracy: 95.25 -> 95.75 | Improved |
| add physics-aware loss (PAL) (ablation) | PATK · FD-B | accuracy: 83.55 -> 84.45 | Improved |
| add LA re-weighting over PA only (ablation) | PATK · EMG | accuracy: 80.38 -> 93.35 | Improved |
| plug PATK tokens into OFA | OFA · Epilepsy | accuracy: 74.13 -> 84.12 | Improved |
| plug PATK tokens into Moirai | Moirai · Epilepsy | accuracy: 74.53 -> 84.78 | Improved |
| plug PATK tokens into CALF | CALF · Epilepsy | accuracy: 75.12 -> 78.34 | Improved |
| plug PATK tokens into OFA for forecasting | OFA · Electricity | MSE: 0.2634 -> 0.2539 | Improved |
| use PATK tokenization versus SimMTM at long horizon | SimMTM · ETTh1 horizon 720 | MSE: 0.4468 -> 0.4402 | Improved |
| use PATK on standard-length forecasting | PATK · ETTh1 horizon 96 | MSE: 0.3792 vs 0.3715 (VisionTS) | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- VisionTS — VisionTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters (ICML 2025) · better on 4 of 5 shared results
- SimMTM — SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling (NeurIPS 2023) · better on 5 of 5 shared results
- TSLANet — TSLANet: Rethinking Transformers for Time Series Representation Learning (ICML 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Enhancing Foundation Models for Time Series Forecasting via Wavelet-based Tokenization (ICML 2025 · similarity 0.88)
- $S^2$IP-LLM: Semantic Space Informed Prompt Learning with LLM for Time Series Forecasting (ICML 2024 · similarity 0.87)
- TimeTok: Granularity-Controllable Time-Series Generation via Hierarchical Tokenization (NeurIPS 2026 · similarity 0.87)
- Kairos: Toward Adaptive and Parameter-Efficient Time Series Foundation Models (NeurIPS 2026 · similarity 0.87)
- Time Series Generation Under Data Scarcity: A Unified Generative Modeling Approach (NeurIPS 2025 · similarity 0.86)
- Multi-Patch Prediction: Adapting Language Models for Time Series Representation Learning (ICML 2024 · similarity 0.86)
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.
