Pioneer: Physics-informed Riemannian Graph ODE for Entropy-increasing Dynamics
Authors: Li Sun, Ziheng Zhang, Zixi Wang, Yujie Wang, Qiqi Wan, Hao Li, Hao Peng, Philip S. Yu
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results show the superiority of Pioneer on real datasets. Extensive empirical results show Pioneer outperforms the state-of-the-art methods on several real datasets, and we analyze the geometry and entropy of real systems. The experiments are conducted on three real-world datasets of dynamic systems, including Weather (Nikitin et al. 2022), CRe SIS (Gogineni et al. 2013) and Social (Gu, Sun, and Gao 2017). To evaluate Poineer, we compare with nine strong baselines of two categories: (1) RNN-based methods... (2) ODE-based methods... We focus on predicting the future trajectories, and evaluate all methods by the metrics of Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) following (Huang, Sun, and Wang 2021; Luo et al. 2023). We perform 10 independent runs for each case, and report the mean with standard derivations. Ablation Study We evaluate the effectiveness of the proposed 1) constrained Ricci flow, and 2) manifold-preserving Gyro-transform. |
| Researcher Affiliation | Academia | 1North China Electric Power University, Beijing 102206, China 2Beihang University, Beijing 100191, China 3Department of Computer Science, University of Illinois at Chicago, IL 60607, USA EMAIL; EMAIL; EMAIL |
| Pseudocode | Yes | Algorithm 1: Learning Algorithm of Pioneer |
| Open Source Code | Yes | Code https://github.com/nakks2/Pioneer |
| Open Datasets | Yes | The experiments are conducted on three real-world datasets of dynamic systems, including Weather (Nikitin et al. 2022), CRe SIS (Gogineni et al. 2013) and Social (Gu, Sun, and Gao 2017). |
| Dataset Splits | No | The paper mentions evaluating predictions over varying lengths (e.g., "prediction length of 3 means we examine the prediction error over the future 3 years") and mentions preprocessing following another paper, but does not explicitly state the specific train/test/validation split percentages, sample counts, or a detailed methodology for splitting the datasets. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using the "Euler method on the manifold (Bielecki 2002)" and optimizing parameters with "Adam with learning rate of 5e-4", but it does not specify version numbers for any software libraries, programming languages, or other tools used. |
| Experiment Setup | Yes | In Pioneer, graph attention layer in manifold ODE is stacked twice, and MLP has 1 hidden layers. The dimension of latent representation on manifold S is set as 16. For Euclidean features, we apply the exponential map to obtain corresponding manifold features in the initialization. The differential system is solved by Euler method on the manifold (Bielecki 2002), and parameters are optimized by Adam with learning rate of 5e-4. |