GREED: A Neural Framework for Learning Graph Distance Functions

Authors: Rishabh Ranjan, Siddharth Grover, Sourav Medya, Venkatesan Chakaravarthy, Yogish Sabharwal, Sayan Ranu

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we establish the following: Efficacy: GREED is more accurate than the state of the art approaches for both GED and SED. Efficiency: GREED is orders of magnitude faster than existing approaches and scales well to graphs with millions of nodes. Scalability: Pair-independence and indexability further enhances the scalability of GREED and enables it to be run on CPU-only platforms. Our code base and datasets are available at https://github.com/idea-iitd/greed. 4.1 Experimental Setup
Researcher Affiliation Collaboration Rishabh Ranjan, Siddharth Grover Department of Computer Science & Engineering, IIT Delhi, India {rishabh.ranjan.cs118, siddharth.grover.cs118}@cse.iitd.ac.in Sourav Medya Department of Computer Science University of Illinois, Chicago, USA medya@uic.edu Venkat Chakravarthy, Yogish Sabharwal IBM Research Delhi, India {vechakra, ysabharwal}@in.ibm.com Sayan Ranu Department of Computer Science & Engineering and Yardi School of AI (Jointly), IIT Delhi, India sayanranu@cse.iitd.ac.in
Pseudocode No The paper describes the architecture and mathematical formulations but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code Yes Our code base and datasets are available at https://github.com/idea-iitd/greed.
Open Datasets Yes Datasets: Table A lists the datasets used for benchmarking. Further details on the dataset semantics are provided in the App. E. We include a mixture of both graph databases (#graphs >1), as well as single large graphs (#graphs = 1). Linux and IMDB are unlabeled. We note that this is the first study to evaluate neural graph distance approaches on million-scale graphs. ... Our code base and datasets are available at https://github.com/idea-iitd/greed.
Dataset Splits Yes Train-Validation-Test: We use 100K query-target pairs for training and 10K pairs each for validation and test.
Hardware Specification Yes We use a machine with an Intel Xeon Gold 6142 processor and Ge Force GTX 1080 Ti GPU for all our experiments.
Software Dependencies No The paper mentions using GIN and GEDLIB implementations but does not provide specific version numbers for any software dependencies or libraries.
Experiment Setup Yes For GREED, we set the number of layers in GIN to 8. The hidden layer dimension is set to 64. For all baselines, we use the default parameters suggested by the authors.