Hierarchical Sliced Wasserstein Distance
Authors: Khai Nguyen, Tongzheng Ren, Huy Nguyen, Litu Rout, Tan Minh Nguyen, Nhat Ho
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we compare the computational cost and generative quality of HSW with the conventional SW on the task of deep generative modeling using various benchmark datasets including CIFAR10, Celeb A, and Tiny Image Net. |
| Researcher Affiliation | Academia | Khai Nguyen Department of Statistics and Data Sciences The University of Texas at Austin Austin, TX 78712 khainb@utexas.eduTongzheng Ren Department of Computer Science The University of Texas at Austin Austin, TX 78712 tongzheng@utexas.eduHuy Nguyen Department of Statistics and Data Sciences The University of Texas at Austin Austin, TX 78712 huynm@utexas.eduLitu Rout Department of Electrical and Computer Engineering The University of Texas at Austin Austin, TX 78712 litu.rout@utexas.eduTan Nguyen Department of Mathematics University of California, Los Angeles Los Angeles, CA 90095 tanmnguyen89@ucla.eduNhat Ho Department of Statistics and Data Sciences The University of Texas at Austin Austin, TX 78712 minhnhat@utexas.edu |
| Pseudocode | Yes | Algorithm 1 Max sliced Wasserstein distance; Algorithm 2 Max hierarchical sliced Wasserstein distance |
| Open Source Code | Yes | Code for experiments in the paper is published at the following link https://github.com/ UT-Austin-Data-Science-Group/HSW. |
| Open Datasets | Yes | standard image datasets including CIFAR10, Celeb A, and Tiny Image Net |
| Dataset Splits | No | The paper mentions training, mini-batch size, and evaluation metrics but does not explicitly specify a validation dataset split or strategy, apart from using standard benchmarks which often imply pre-defined splits without explicitly stating them here. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory used for running the experiments. It only generally refers to 'deep learning applications'. |
| Software Dependencies | No | The paper mentions using the Adam optimizer but does not specify its version number or versions for other software components like Python or PyTorch. |
| Experiment Setup | Yes | For all datasets, the number of training iterations is set to 50000. We update the generator Gϕ for each 5 iterations using (Max-)SW and (Max-)HSW. Moreover, we update Dβ1 and Dβ2 ( the discriminator) every iterations. The mini-batch size m is set 128 in all datasets. We set the learning rate for Gϕ, Dβ1, and Dβ2 to 0.0002. The optimizer that we use is Adam (Kingma & Ba, 2014) with parameters (β1, β2) = (0, 0.9) (slightly abuse of notations). For SW and HSW, we use p = 2 (the order of Wasserstein distance). |