NuwaDynamics: Discovering and Updating in Causal Spatio-Temporal Modeling

Authors: Kun Wang, Hao Wu, Yifan Duan, Guibin Zhang, Kai Wang, Xiaojiang Peng, Yu Zheng, Yuxuan Liang, Yang Wang

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on six real-world ST benchmarks showcase that models can gain outcomes upon the integration of the Nuwa Dynamics concept. Nuwa Dynamics also can significantly benefit a wide range of changeable ST tasks like extreme weather and long temporal step super-resolution predictions.
Researcher Affiliation Collaboration Kun Wang2, , Hao Wu4, , Yifan Duan3, Guibin Zhang6, Kai Wang8, Xiaojiang Peng7, Yu Zheng9, Yuxuan Liang5 , Yang Wang1,2,3,4 1 Key Laboratory of Precision and Intelligent Chemistry, University of Science and Technology of China (USTC) 2Suzhou Institute for Advanced Research, USTC 3School of Software Engineering, USTC 4 School of Computer Science, USTC 5 Hong Kong University of Science and Technology (Guangzhou) 6 Tongji University 7 Shenzhen Technology University 8 National University of Singapore 9 JD i City, JD Technology
Pseudocode No The paper describes its methodology using textual descriptions and diagrams (e.g., Figure 3), but it does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Our codes are available at https://github.com/easylearningscores/Nuwa Dynamics.
Open Datasets Yes Datasets & Backbones We extensively evaluate our proposal on five benchmarks across diverse research domains, including Taxi BJ+ (Liang et al., 2021), KTH (Schuldt et al., 2004), SEVIR (Veillette et al., 2020), Rain Net (Ayzel et al., 2020), PD, and Fire Sys (Chen et al., 2022a).
Dataset Splits No Appendix C, Table 4, states: 'N tr and N te denote the number of instances in the training and test sets.' However, the paper does not explicitly provide details on how validation sets were created or their specific sizes/percentages, nor does it refer to predefined validation splits with citations.
Hardware Specification Yes We implement different backbones using Pytorch and leveraging the A100-PCIE40GB as support.
Software Dependencies No The paper mentions implementing backbones 'using Pytorch', but it does not specify a version number for Pytorch or any other software dependencies, which is necessary for reproducibility.
Experiment Setup Yes We train all models with Adam optimizer and learning rate as 0.01.