Patch-wise Structural Loss for Time Series Forecasting
Authors: Dilfira Kudrat, Zongxia Xie, Yanru Sun, Tianyu Jia, Qinghua Hu
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that PS loss significantly improves the performance of state-of-the-art models across diverse real-world datasets. The paper includes Section 4 titled "Experiments" with subsections detailing experimental setup, main results, comparisons, ablation studies, and visualizations. |
| Researcher Affiliation | Academia | All authors are affiliated with "1College of Intelligence and Computing, Tianjin University, China." |
| Pseudocode | No | The paper describes the methodology using prose and mathematical formulas, but does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The data and code are publicly available at: https://github. com/Dilfiraa/PS_Loss. |
| Open Datasets | Yes | We conduct our experiments on seven real-world multivariate time series datasets, including ETT (ETTh1, ETTh2, ETTm1, ETTm2), Weather, ECL, and Exchange. For data preprocessing, we follow the standard protocol used by backbone studies. The detailed description of the datasets is provided in Appendix A. ETT (Electricity Transformer Temperature): The ETT (Zhou et al., 2021) dataset... Weather: The Weather (Zhou et al., 2021) dataset... ECL (Electricity Consuming Load): The ECL (Li et al., 2019) dataset... Exchange: The Exchange (Lai et al., 2018) dataset... |
| Dataset Splits | Yes | The ETT dataset is divided into training, validation, and testing sets in a 12/4/4-month ratio, while other datasets are split in a 7:1:2 ratio. |
| Hardware Specification | Yes | The experiments are implemented using Py Torch and executed on NVIDIA RTX 3090 24GB GPU. |
| Software Dependencies | No | The experiments are implemented using Py Torch and executed on NVIDIA RTX 3090 24GB GPU. While PyTorch is mentioned, no specific version number for PyTorch or other software dependencies is provided. |
| Experiment Setup | Yes | For fair evaluation, when integrating PS loss to enhance the backbone model s performance, we follow their original experimental and hyperparameter settings, while only tuning the PS loss weight λ and the patch length threshold δ. ... We evaluate a range of λ {0.1, 0.3, 0.5, 0.7, 1.0, 3.0, 5.0, 10.0} ... We evaluate δ {24, 36, 48, 60}... |