Distributional Sliced-Wasserstein and Applications to Generative Modeling

Authors: Khai Nguyen, Nhat Ho, Tung Pham, Hung Bui

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct extensive experiments with large-scale datasets to demonstrate the favorable performances of the proposed distances over the previous sliced-based distances in generative modeling applications.
Researcher Affiliation Collaboration Khai Nguyen Vin AI Research, Vietnam v.khainb@vinai.io Nhat Ho University of Texas, Austin Vin AI Research, Vietnam minhnhat@utexas.edu Tung Pham Vin AI Research, Vietnam v.tungph4@vinai.io Hung Bui Vin AI Research, Vietnam v.hungbh1@vinai.io
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link to its own open-source code for the methodology described.
Open Datasets Yes We conduct extensive experiments ... on MNIST (Le Cun et al., 1998), CIFAR10 (Krizhevsky, 2009), Celeb A (Liu et al., 2015) and LSUN (Yu et al., 2015) datasets.
Dataset Splits Yes For λC in DSW (see Definition 2), it is chosen in the set {1, 10, 100, 1000} such that its Wasserstein-2 (FID score) (between 10000 random generated images and all images from corresponding validation set) is the lowest among the four values.
Hardware Specification No The paper mentions 'computational time' but does not provide any specific details about the hardware (e.g., CPU, GPU models, memory) used for the experiments.
Software Dependencies No The paper mentions external libraries like 'Pot python optimal transport library' and 'Geoopt: Riemannian optimization in pytorch' in its references, but it does not specify version numbers for these or any other software dependencies within the main text or experimental setup.
Experiment Setup Yes Detailed experiment settings are in Appendix G. ... We set the minibatch size be 512 on Celeb A and CIFAR, and be 4096 on LSUN. ... For λC in DSW (see Definition 2), it is chosen in the set {1, 10, 100, 1000}... We let N, the number of projections of both DSW and SW, vary in the set {1, 10, 102, 5 102, 103, 5 103, 104} for the SW, and N {1, 10, 102, 5 102, 103, 5 103} for the DSW.