Outlier-Robust Gromov-Wasserstein for Graph Data

Authors: Lemin Kong, Jiajin Li, Jianheng Tang, Anthony Man-Cho So

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experimentation, we validate our theoretical results and demonstrate the effectiveness of RGW on real-world graph learning tasks, such as subgraph matching and partial shape correspondence.
Researcher Affiliation Academia Lemin Kong CUHK lkong@se.cuhk.edu.hk Jiajin Li Stanford University jiajinli@stanford.edu Jianheng Tang HKUST jtangbf@connect.ust.hk Anthony Man-Cho So CUHK manchoso@se.cuhk.edu.hk
Pseudocode No The paper describes algorithms (BPALM) using mathematical equations and textual explanations, but it does not include an explicitly labeled "Pseudocode" or "Algorithm" block.
Open Source Code Yes Our code is available at https://github.com/lmkong020/outlier-robust-GW.
Open Datasets Yes We evaluate the matching performance of RGW on the TOSCA dataset [6, 34] for partial shape correspondence. [...] The Proteins and Enzymes biological graph databases from [15] are also used [...] Additionally, we also evaluate our methods on the Douban Online-Offline social network dataset...
Dataset Splits No The paper describes the datasets used and how source graphs are generated (e.g., "sampling connected subgraphs... with a specified percentage of nodes"), but it does not explicitly specify training, validation, or test dataset splits with percentages, absolute counts, or references to predefined splits for reproducibility.
Hardware Specification Yes All simulations are conducted in Python 3.9 on a high-performance computing server running Ubuntu 20.10, equipped with an Intel(R) Xeon(R) Silver 4214R CPU.
Software Dependencies No The paper mentions "Python 3.9" and "Ubuntu 20.10" but does not list multiple key software libraries, frameworks, or solvers with their specific version numbers (e.g., PyTorch, TensorFlow, specific optimization libraries).
Experiment Setup Yes RGW sets τ1 = τ2 = 0.1 and selects the best results from the ranges {0.05, 0.1, 0.2, 0.5} for marginal relaxation parameters and {0.01, 0.05, 0.1, 0.5, 1} for step size parameters (tk, ck, rk). The transport plan initialization uses different approaches depending on the dataset.