From Stars to Subgraphs: Uplifting Any GNN with Local Structure Awareness
Authors: Lingxiao Zhao, Wei Jin, Leman Akoglu, Neil Shah
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method sets new state-of-the-art performance by large margins for several well-known graph ML tasks; specifically, 0.08 MAE on ZINC, 74.79% and 86.887% accuracy on CIFAR10 and PATTERN respectively. ... We conduct extensive experiments on 4 simulation datasets and 5 well-known real-world graph classification & regression benchmarks (Dwivedi et al., 2020; Hu et al., 2020), to show significant and consistent practical benefits of our approach across different MPNNs and datasets. |
| Researcher Affiliation | Collaboration | Lingxiao Zhao Carnegie Mellon Uni. lingxiao@cmu.edu Wei Jin Michigan State Uni. jinwei2@msu.edu Leman Akoglu Carnegie Mellon Uni. lakoglu@andrew.cmu.edu Neil Shah Snap Inc. nshah@snap.com |
| Pseudocode | No | The paper describes its method using equations and textual explanations, along with diagrams, but does not provide a formal pseudocode block or algorithm listing. |
| Open Source Code | Yes | Our implementation is easy-to-use, and directly accepts any GNN from Py G (Fey & Lenssen, 2019) for plug-and-play use. See code at https://github.com/GNNAs Kernel/GNNAs Kernel. |
| Open Datasets | Yes | Simulation Datasets: 1) EXP (Abboud et al., 2021) ... 2) SR25 (Balcilar et al., 2021) ... Large Real-world Datasets: ZINC-12K, CIFAR10, PATTER from Benchmarking GNNs (Dwivedi et al., 2020) and Mol HIV, and Mol PCBA from Open Graph Benchmark (Hu et al., 2020). Small Real-world Datasets: MUTAG, PTC, PROTEINS, NCI1, IMDB, and REDDIT from TUDatset (Morris et al., 2020a). |
| Dataset Splits | Yes | Table 5: Dataset statistics. ... ZINC-12K ... 10000 / 1000 / 1000 ... CIFAR10 ... 45000 / 5000 / 10000 ... PATTERN ... 10000 / 2000 / 2000 ... Mol HIV ... 32901 / 4113 / 4113 ... Mol PCBA ... 350343 / 43793 / 43793 ... To reduce the search space, we search hyperparameters in a two-phase approach: First, we search common ones (hidden size from [64, 128], number of layers L from [2,4,5,6], (sub)graph pooling from [SUM, MEAN] for each dataset using GIN based on validation performance, and fix it for any other GNN and GNN-AK(+). |
| Hardware Specification | Yes | All experiments are conducted on RTX-A6000 GPUs. |
| Software Dependencies | No | The paper mentions 'directly accepts any GNN from Py G (Fey & Lenssen, 2019)' but does not specify version numbers for PyG or any other software components like Python or PyTorch. |
| Experiment Setup | Yes | To reduce the search space, we search hyperparameters in a two-phase approach: First, we search common ones (hidden size from [64, 128], number of layers L from [2,4,5,6], (sub)graph pooling from [SUM, MEAN] for each dataset using GIN based on validation performance, and fix it for any other GNN and GNN-AK(+). ... We use Batch Normalization and Re LU activation in all models. For optimization we use Adam with learning rate 0.001 and weight decay 0. |