Node Embedding from Neural Hamiltonian Orbits in Graph Neural Networks

Authors: Qiyu Kang, Kai Zhao, Yang Song, Sijie Wang, Wee Peng Tay

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments demonstrate that our approach adapts better to different types of graph datasets than popular state-of-the-art graph node embedding GNNs.
Researcher Affiliation Collaboration 1School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore 2C3 AI, Singapore.
Pseudocode Yes Algorithm 1 Graph Node Embedding with Ham GNN
Open Source Code Yes The code is available at https://github.com/zknus/ Hamiltonian-GNN.
Open Datasets Yes We select datasets with various geometries including the three citation networks: Cora (Mc Callum et al., 2004), Citeseer (Sen et al., 2008), Pubmed (Namata et al., 2012); and two low hyperbolicity datasets (Chami et al., 2019): Disease and Airport (cf. Table 5). and In this section, to underscore our model s capacity for handling large graph datasets, we conduct a series of experiments on the Ogbn datasets obtained from https: //ogb.stanford.edu/docs/nodeprop/, in compliance with the experimental setup detailed in (Hu et al., 2021).
Dataset Splits Yes We use a 60%, 20%, 20% random split for training, validation, and test sets on this new dataset.
Hardware Specification No The paper discusses computational cost and time in Table 11, but does not provide specific hardware details such as CPU/GPU models, memory, or cluster specifications used for the experiments.
Software Dependencies No We employ the ODE solver (Chen, 2018) in the implementation of Ham GNN. For computation efficiency and performance effectiveness, the fixed-step explicit Euler solver (Chen et al., 2018a) is used in Ham GNN.
Experiment Setup Yes We use the ADAM optimizer (Kingma & Ba, 2014) with the weight decay as 0.001. We set the learning rate as 0.01 for citation networks and 0.001 for Disease and Airport datasets. The results presented in Table 1 are under the 3 layers Ham GNN setting. Ham GNN first compresses the dimension of input features to the fixed hidden dimension (e.g. 64) through a fully connected (FC) layer.