Semidefinite Relaxations of the Gromov-Wasserstein Distance

Authors: Junyu Chen, Binh T. Nguyen, Shang Koh, Yong Sheng Soh

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our numerical experiments suggest that the proposed relaxation is strong in that it frequently computes the globally optimal solution. Our Python implementation is available at https://github.com/tbng/gwsdp. ... 4 Numerical Experiments with Off-the-shelf Convex Solvers
Researcher Affiliation Academia Junyu Chen Binh T. Nguyen Shang Hui Koh Yong Sheng Soh Department of Mathematics National University of Singapore chenjunyu@u.nus.edu,binhnt@nus.edu.sg,matsys@nus.edu.sg
Pseudocode Yes Algorithm 1 Computation of GW-SDP barycenters. Input: dataset {Ck, αk}K k=1; {λk}K k=1. Initialize C. repeat for k = 1 to K do πsdp,k solve_GW-SDP(Ck, C, αk, α). end for Update C using (8). until convergence
Open Source Code Yes Our Python implementation is available at https://github.com/tbng/gwsdp.
Open Datasets Yes We use a publicly available dataset of triangular meshes (Sumner and Popovi c, 2004).
Dataset Splits No The paper does not explicitly state specific train/validation/test dataset splits (e.g., percentages or sample counts) for any of its experiments.
Hardware Specification Yes Table 1 presents the run-time of the GW-SDP problem in Experiment 1, running on a PC with 8 cores CPU and 32GB of RAM.
Software Dependencies Yes We solve the GW-SDP instance implemented in CVXPY (Diamond and Boyd, 2016) using the SCS and MOSEK solvers (Ap S, 2022; O Donoghue et al., 2016). ... The MOSEK optimization toolbox for Python manual. Version 10.0.
Experiment Setup No The paper describes the general setup of different numerical experiments (e.g., matching Gaussian distributions, graph community matching) but does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) or system-level training settings for the optimization process itself.