GRAFENNE: Learning on Graphs with Heterogeneous and Dynamic Feature Sets

Authors: Shubham Gupta, Sahil Manchanda, Sayan Ranu, Srikanta J. Bedathur

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical evaluation: Extensive experiments on diverse real-world datasets establish that GRAFENNE consistently outperforms baseline methods across various levels of feature scarcity on both homophilic as well as heterophilic graphs.
Researcher Affiliation Academia 1Department of Computer Science and Engineering, IIT Delhi.
Pseudocode No The paper describes the message passing layers using mathematical equations and text, but does not include a formal pseudocode block or algorithm listing.
Open Source Code Yes Our codebase is available at https://github.com/data-iitd/Grafenne.
Open Datasets Yes We evaluate GRAFENNE on the real-world graphs listed in Table 1. Among these, Actor is a heterophilic graph, whereas the rest are homophilic. Further details on the semantics of the datasets are provided in App. H. Table 1: Dataset statistics Cora (Sen et al., 2008) Cite Seer (Yang et al., 2016) Physics (Shchur et al., 2018) Actor (Pei et al., 2020)
Dataset Splits Yes We perform a 60% 20% 20% data split for train-test-validation.
Hardware Specification Yes All experiments are performed on an Intel Xeon Gold 6248 processor with 80 cores, 1 Tesla V-100 GPU card with 32GB GPU memory, and 377 GB RAM with Ubuntu 18.04.
Software Dependencies No The paper specifies the operating system (Ubuntu 18.04) but does not list specific versions for other key software components, libraries, or frameworks used (e.g., Python, PyTorch, TensorFlow).
Experiment Setup Yes We have used 2 layers of message-passing and trained GRAFENNE using the Adam optimizer with a learning rate of 0.0001 and choose the model based on the best validation loss.