Graph As Point Set

Authors: Xiyuan Wang, Pan Li, Muhan Zhang

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

Reproducibility Variable Result LLM Response
Research Type Experimental In our experiments, we evaluate our model across three dimensions: substructure counting for short-range expressivity, real-world graph property prediction for practical performance, and Long-Range Graph Benchmarks (Dwivedi et al., 2022b) to assess long-range interactions.
Researcher Affiliation Academia 1Institute for Artificial Intelligence, Peking University 2Georgia Institute of Technology.
Pseudocode No The paper describes its methods in text and uses figures to illustrate architectures (e.g., Figure 1, Figure 3, Figure 4), but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Our code is available at https://github.com/GraphPKU/GraphAsSet.
Open Datasets Yes Synthetic is the dataset used in substructure counting tasks provided by Huang et al. (2023b), they are random graph with the count of substructure as node label. QM9 (Wu et al., 2017), ZINC (G omez Bombarelli et al., 2016), and ogbg-molhiv are three datasets of molecules.
Dataset Splits Yes In split column, fixed means the dataset uses the split provided in the original release. Otherwise, it is of the formal training set ratio/valid ratio/test ratio. ... Synthetic 0.3/0.2/0.5. QM9 0.8/0.1/0.1 ... MUTAG ... 10-fold cross validataion
Hardware Specification Yes All our experiments are conducted on NVIDIA RTX 3090 GPUs on a linux server.
Software Dependencies No Our code is primarily based on Py Torch (Paszke et al., 2019) and Py G (Fey & Lenssen, 2019). (Specific version numbers for PyTorch and PyG are not provided.)
Experiment Setup Yes We use l1 loss for regression tasks and cross entropy loss for classification tasks... For optimization, we use Adam W optimizer and cosine annealing scheduler. Hyperparameters for datasets are shown in Table 7.