KITS: Inductive Spatio-Temporal Kriging with Increment Training Strategy
Authors: Qianxiong Xu, Cheng Long, Ziyue Li, Sijie Ruan, Rui Zhao, Zhishuai Li
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments demonstrate that KITS consistently outperforms existing methods by large margins, e.g., the improvement over MAE score could be as high as 18.33%. We conduct experiments on eight datasets of three types, showing that our KITS outperforms existing methods consistently by large margins (as high as 18.33%). |
| Researcher Affiliation | Collaboration | Qianxiong Xu1, Cheng Long1*, Ziyue Li2*, Sijie Ruan3, Rui Zhao4, Zhishuai Li4 1S-Lab, Nanyang Technological University 2University of Cologne 3Beijing Institute of Technology 4Sense Time Research |
| Pseudocode | Yes | Note that the pseudo code of Increment training strategy is included in Appendix. |
| Open Source Code | Yes | Code https://github.com/Sam1224/KITS |
| Open Datasets | Yes | We employ 8 public datasets and conduct extensive experiments on them, so as to validate the effectiveness of KITS. These datasets are collected from different real-world application scenarios, including 4 datasets in the field of traffic (METR-LA, PEMS-BAY, SEA-LOOP, PEMS07), 2 in air quality (AQI-36, AQI), and 2 in solar power (NRELAL, NREL-MD). |
| Dataset Splits | Yes | We set the missing ratios α = 50% for all datasets and present the results in Table 1 for inductive setting, and in Table 2 for transductive setting. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., GPU/CPU models, memory amounts) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details, such as library names with version numbers, needed to replicate the experiment. |
| Experiment Setup | No | The paper mentions that "λ is a hyperparameter controlling the importance of pseudo labels" and discusses the probability `p` for creating edges for virtual nodes, but it does not provide comprehensive specific experimental setup details such as learning rates, batch sizes, number of epochs, or optimizer settings in the main text. |