ASAP: Adaptive Structure Aware Pooling for Learning Hierarchical Graph Representations

Authors: Ekagra Ranjan, Soumya Sanyal, Partha Talukdar5470-5477

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments on multiple datasets and theoretical analysis, we motivate our choice of the components used in ASAP. Our experimental results show that combining existing GNN architectures with ASAP leads to state-of-the-art results on multiple graph classification benchmarks.
Researcher Affiliation Academia 1Indian Institute of Technology, Guwahati 2Indian Institute of Science, Bangalore
Pseudocode Yes Please refer to Appendix Sec. I for a pseudo code of the working of ASAP.
Open Source Code Yes We make the source code of ASAP available to encourage reproducible research 1. 1https://github.com/malllabiisc/ASAP
Open Datasets Yes D&D (Shervashidze et al. 2011; Dobson and Doig 2003) and PROTEINS (Dobson and Doig 2003; Borgwardt et al. 2005) are datasets containing proteins as graphs. NCI1 (Wale, Watson, and Karypis 2008) and NCI109 are datasets for anticancer activity classification. FRANKENSTEIN (Orsini, Frasconi, and De Raedt 2015) contains molecules as graph for mutagen classification.
Dataset Splits Yes Following SAGPool(Lee, Lee, and Kang 2019), we conduct our experiments using 10-fold cross-validation and report the average accuracy on 20 random seeds.
Hardware Specification No The paper does not explicitly describe the hardware used for running its experiments, such as specific GPU or CPU models.
Software Dependencies No The paper does not provide specific version numbers for key software components or libraries used for the experiments.
Experiment Setup Yes For ASAP, we choose k = 0.5 and h = 1 to be consistent with baselines. Following SAGPool(Lee, Lee, and Kang 2019), we conduct our experiments using 10-fold cross-validation and report the average accuracy on 20 random seeds. Please refer to Appendix Sec. A for further details on hyperparameter tuning.