Gromov-Wasserstein Autoencoders

Authors: Nao Nakagawa, Ren Togo, Takahiro Ogawa, Miki Haseyama

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

Reproducibility Variable Result LLM Response
Research Type Experimental The empirical comparisons with VAE-based models show that GWAE models work in two prominent meta-priors, disentanglement and clustering, with their GW objective unchanged. We conduct empirical evaluations on the capability of GWAE in prominent meta-priors: disentanglement and clustering. Several experiments on image datasets Celeb A (Liu et al., 2015), MNIST (Le Cun et al., 1998), and 3D Shapes (Burgess & Kim, 2018), show that GWAE models outperform the VAE-based representation learning methods whereas their GW objective is not changed over different meta-priors.
Researcher Affiliation Academia Graduate School of Information Science and Technology, Hokkaido University, Japan Faculty of Information Science and Technology, Hokkaido University, Japan
Pseudocode No The paper describes its methods through text and mathematical formulations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes To ensure reproducibility, our code is available online at https://github.com/ganmodokix/gwae and is provided as the supplementary material.
Open Datasets Yes For the reported experiments in Section 4, we used the following datasets: MNIST (Le Cun et al., 1998). Celeb A (Liu et al., 2015). 3D Shapes (Burgess & Kim, 2018). Omniglot (Lake et al., 2015). CIFAR-10 (Krizhevsky & Hinton, 2009).
Dataset Splits Yes We used the original test set and randomly split the original training set into 54,000 training images and 6,000 validation images. (for MNIST) We randomly split the entire dataset into 384,000/48,000/48,000 images for the train/validation/test set, respectively. (for 3D Shapes)
Hardware Specification Yes For the reported experimental results, we used a single GPU of NVIDIA Ge Force RTX 2080 Ti, and a single run of the entire GWAE training process until convergence takes about eight hours.
Software Dependencies No The paper mentions 'Py Torch (Paszke et al., 2019)' as a framework and optimizers like 'RMSProp' and 'Adam (Kingma & Ba, 2015)', but does not provide specific version numbers for these software components or any other libraries used.
Experiment Setup Yes For quantitative evaluations, we selected hyperparameters from λW [100, 101], λD [100, 101], and λH [10 4, 100] using their performance on the validation set. For the optimizers of GWAE, we used RMSProp with a learning rate of 10 4 for the main autoencoder network and used RMSProp with a learning rate of 5 10 5 for the critic network. For all the compared methods except for GWAE, we used the Adam (Kingma & Ba, 2015) optimizer with a learning rate of 10 4. In the experiments, we used an equal batch size of 64 for all evaluated models.