Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Augmented Sliced Wasserstein Distances
Authors: Xiongjie Chen, Yongxin Yang, Yunpeng Li
ICLR 2022 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical results demonstrate that the ASWD significantly outperforms other Wasserstein variants for both synthetic and real-world problems. ... In this section, we describe the experiments that we have conducted to evaluate performance of the proposed distance metric. |
| Researcher Affiliation | Collaboration | 1University of Surrey, 2University of Edinburgh, 3Huawei Noah s Ark Lab |
| Pseudocode | Yes | Pseudocode is provided in Appendix E. (Algorithm 1 The augmented sliced Wasserstein distance.) |
| Open Source Code | Yes | Code to reproduce experiment results is available at : https://github.com/xiongjiechen/ASWD. |
| Open Datasets | Yes | CIFAR 10 (Krizhevsky, 2009), Celeb A (Liu et al., 2015), and MNIST (Le Cun et al., 1998) datasets |
| Dataset Splits | No | The paper mentions using well-known datasets like CIFAR10 and Celeb A for training and evaluation but does not explicitly state the specific training/validation/test splits (e.g., percentages or counts) within the text. |
| Hardware Specification | Yes | The running time per training iteration for one batch containing 512 samples is computed based on a computer with an Intel (R) Xeon (R) Gold 5218 CPU 2.3 GHz and 16GB of RAM, and a RTX 6000 graphic card with 22GB memories. |
| Software Dependencies | No | The paper mentions using the Adam optimizer and specific activation functions (ReLU, Tanh, Sigmoid) but does not provide version numbers for any software libraries, frameworks (e.g., PyTorch, TensorFlow), or programming languages used. |
| Experiment Setup | Yes | We initialize the source distributions ยต0 as standard normal distributions N(0,I). We update source distributions using Adam optimizer, and set the learning rate=0.002. For all methods, we set the order k = 2. When testing the ASWD, the number of iterations M in Algorithm 1 is set to 10. ... We train the models with the Adam optimizer, and set the batch size to 512. Following the setup in (Nguyen et al., 2021), the learning rate is set to 0.0005 and beta=(0.5, 0.999) for both CIFAR10 dataset and Celeb A dataset. For the ASWD, the number of iterations M in Algorithm 1 is set to 5. The hyperparameter ฮป is set to 1.01. |