Gromov-Wasserstein Learning for Graph Matching and Node Embedding

Authors: Hongteng Xu, Dixin Luo, Hongyuan Zha, Lawrence Carin Duke

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the Gromov-Wasserstein learning (GWL) method to both synthetic and real-world matching tasks, and compare it with state-of-the-art methods. In our experiments, we set hyperparameters as follows: the number of outer iterations is M = 30, the number of inner iteration is N = 200, γ = 0.01 and L( , ) is the MSE loss.
Researcher Affiliation Collaboration 1Infinia ML, Inc., Durham, NC, USA 2Department of ECE, Duke University, Durham, NC, USA 3College of Computing, Georgia Institute of Technology, Atlanta, GA, USA.
Pseudocode Yes Algorithm 1 Gromov-Wasserstein Learning (GWL)
Open Source Code Yes The code is available on https://github.com/Hongteng Xu/gwl.
Open Datasets Yes MC3 is a dataset used in the Mini-Challenge 3 of VAST Challenge 2018, which records the communication behavior among a company s employees on different networks.1 The communications are categorized into two types: phone calls and emails between employees. ... 1http://vacommunity.org/VAST+Challenge+2018+MC3
Dataset Splits Yes For all the methods, we use 50% of the admissions for training, 25% for validation, and the remaining 25% for testing.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments.
Software Dependencies No The paper mentions using Adam as an optimizer, but does not provide specific version numbers for any software dependencies or libraries (e.g., Python, deep learning frameworks, or numerical libraries) used in the experiments.
Experiment Setup Yes In our experiments, we set hyperparameters as follows: the number of outer iterations is M = 30, the number of inner iteration is N = 200, γ = 0.01 and L( , ) is the MSE loss. We tried βs in {0, 1, 10, 100, 1000} and the β in [1, 100] achieves stable performance. Therefore, we empirically set β = 10. When solving (8), we use Adam (Kingma & Ba, 2014) with learning rate 0.001 and set the number of epochs to 5, and the size of batches as 100.