Continuous Product Graph Neural Networks
Authors: Aref Einizade, Fragkiskos Malliaros, Jhony H. Giraldo
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct thorough theoretical analyses of the stability and over-smoothing properties of CITRUS in response to domain-specific graph perturbations and graph spectra effects on the performance. We evaluate CITRUS on well-known traffic and weather spatiotemporal forecasting datasets, demonstrating superior performance over existing approaches. |
| Researcher Affiliation | Academia | Aref Einizade LTCI, Télécom Paris Institut Polytechnique de Paris aref.einizade@telecom-paris.fr Fragkiskos D. Malliaros Centrale Supélec, Inria Université Paris-Saclay fragkiskos.malliaros@centralesupelec.fr Jhony H. Giraldo LTCI, Télécom Paris Institut Polytechnique de Paris jhony.giraldo@telecom-paris.fr |
| Pseudocode | No | The paper describes algorithms and methods in textual and mathematical forms, but it does not include explicitly labeled pseudocode blocks or algorithm figures. |
| Open Source Code | Yes | The implementation codes are available at https://github.com/Aref Einizade2/CITRUS. |
| Open Datasets | Yes | We evaluate CITRUS on well-known traffic and weather spatiotemporal forecasting datasets: Metr LA [35] and Pems Bay [36]... Molene [43] and NOAA [44]. |
| Dataset Splits | Yes | We utilize 15% of the nodes in the product graph as the test set, 15% of the remaining nodes as validation, and the rest of the nodes for training. |
| Hardware Specification | Yes | All the experiments were run on one GTX A100 GPU device with 40 GB of RAM. |
| Software Dependencies | No | The paper does not provide specific version numbers for software dependencies (e.g., Python, PyTorch, or other libraries). |
| Experiment Setup | Yes | All hyperparameters were optimized on the validation set with the details in Section I. The detailed hyperparameters (optimized by cross-validation on the validation data) and/or training settings of CITRUS , i.e., T (auto-regressive order), emb (dimension of embedding size in the spatiotemporal encoder), hid (dimension of linear mapping size in the spatiotemporal encoder), FMLP (dimension of linear mapping size in the MLP layers), n CITRUS (number of CITRUS blocks), F (second dimension of Wl in (6)), ki (number of selected eigenvalue-eigenvector pairs in the i-th factor graph), lr (learning rate), nbatch (batch size), and nepochs (number of epochs), are provided in Table 9. |