Spherical Sliced-Wasserstein
Authors: Clément Bonet, Paul Berg, Nicolas Courty, François Septier, Lucas Drumetz, Minh Tan Pham
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Then, we show that we can use this discrepancy on different tasks such as sampling, density estimation or generative modeling. We first illustrate the ability to approximate different distributions by minimizing SSW w.r.t. some target distributions on S2 and by performing density estimation experiments on real earth data. Then, we apply SSW for generative modeling tasks using the framework of Sliced-Wasserstein autoencoder and we show that we obtain competitive results with other Wasserstein autoencoder based methods using a prior on higher dimensional hyperspheres. |
| Researcher Affiliation | Academia | Cl ement Bonet1, Paul Berg2, Nicolas Courty2, Franc ois Septier1, Lucas Drumetz3, Minh-Tan Pham2 Universit e Bretagne Sud, LMBA1, IRISA2; IMT Atlantique, Lab-STICC3 {clement.bonet, paul.berg, francois.septier}@univ-ubs.fr {nicolas.courty, minh-tan.pham}@irisa.fr; lucas.drumetz@imt-atlantique.fr |
| Pseudocode | Yes | Algorithm 1 SSW |
| Open Source Code | Yes | The code is available online1. 1https://github.com/clbonet/Spherical_Sliced-Wasserstein |
| Open Datasets | Yes | In the following, we use the MNIST (Le Cun & Cortes, 2010), Fashion MNIST (Xiao et al., 2017) and CIFAR10 (Krizhevsky, 2009) datasets... We perform density estimation on datasets first gathered by Mathieu & Nickel (2020) which contain locations of wild fires (EOSDIS, 2020), floods (Brakenridge, 2017) or eathquakes (NOAA, 2022). |
| Dataset Splits | No | The paper mentions 'split of the data' and provides 'Train set size' and 'Test set size' in an appendix table, but does not explicitly provide validation set splits or percentages in the main text necessary for reproduction. |
| Hardware Specification | No | All experiments are conducted on GPU. This work was performed partly using HPC resources from GENCI-IDRIS (Grant 2022-AD011013514). |
| Software Dependencies | No | Using Pytorch (Paszke et al., 2019), we implemented the binary search algorithm of (Delon et al., 2010) and used it with ϵ = 10 6. We use the Python Optimal Transport (POT) library (Flamary et al., 2021) to compute the Wasserstein distance and the entropic approximation. The paper mentions software names with citations but does not list specific version numbers for reproducibility. |
| Experiment Setup | Yes | Using Pytorch (Paszke et al., 2019), we implemented the binary search algorithm of (Delon et al., 2010) and used it with ϵ = 10 6. We perform a gradient descent using Adam (Kingma & Ba, 2014) as the optimizer with a learning rate of 10 4 for 2000 epochs. For both the encoder and the decoder architecture, we use fully convolutional architectures with 3x3 convolutional filters...Adam as optimizer with a learning rate of 10 3 and Pytorch s default momentum parameters for 800 epochs with batch of size n = 500. |