Towards Better Spherical Sliced-Wasserstein Distance Learning with Data-Adaptive Discriminative Projection Direction
Authors: Hongliang Zhang, Shuo Chen, Lei Luo, Jian Yang
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we evaluate the performance of our proposed DSSW by comparing it with several state-of-the-art methods across a variety of machine learning tasks, including gradient flows, density estimation on real earth data, and self-supervised learning. ... In line with previous works (Bonet et al. 2023; Tran et al. 2024), we conducted five different numerical experiments to validate the effectiveness of our method in comparison to SW, SSW (Bonet et al. 2023) and S3W distance (Tran et al. 2024)... |
| Researcher Affiliation | Academia | 1PCA Lab, Key Lab of Intelligent Perception and Systems for High-Dimensional Information of Ministry of Education, School of Computer Science and Engineering, Nanjing University of Science and Technology, China 2School of Intelligence Science and Technology, Nanjing University, China EMAIL, EMAIL |
| Pseudocode | Yes | The pseudocode for computing the DSSW distance is provided in Algorithms 1 and 2 in Appendix Section B. |
| Open Source Code | No | The paper does not contain any explicit statement about releasing source code or provide a link to a code repository. |
| Open Datasets | Yes | In alignment with (Bonet et al. 2023; Tran et al. 2024), we adopt three datasets (Mathieu and Nickel 2020): Earthquake (NOAA 2022), Flood (Brakenridge 2017) and Fire (EOSDIS 2020). ... The results on the CIFAR10 (Krizhevsky and Hinton 2009) benchmark for SWAE with v MF prior are shown in Table 3. ... additional results and the latent space visualization on the MNIST (Lecun et al. 1998) benchmark in Appendix Section C.6. |
| Dataset Splits | No | The paper mentions using well-known datasets and refers to 'mini-batch results', 'test data', and 'training runs', but does not specify the exact percentages or sample counts for training, validation, or test splits. It implicitly uses standard splits but does not detail them. |
| Hardware Specification | Yes | All our experiments are implemented by PyTorch (Paszke et al. 2019) on Ubuntu 20.04 and a single NVIDIA RTX 4090 GPU. |
| Software Dependencies | No | The paper mentions 'PyTorch (Paszke et al. 2019) on Ubuntu 20.04' but does not specify the version number for PyTorch or any other key software libraries. |
| Experiment Setup | Yes | x i,k+1 =xi,k γ xi,k DSSW (ˆµk, ˆνni) ... where γ is the learning rate for the update rule... The training objective of the revised SWAE is min α,β Ex µ [c (x, α (β (x)))]+η DSSW 2 2 (α#µ, z) , (14) ... where η denotes the regularization coefficient. ... DSSW-SSL = 1 i=1 z A i z B i 2 2 2 DSSW 2 2 (z A, ν)+DSSW 2 2 (z B, ν) , where ... η > 0 acts as a regularization coefficient... Tables 3 and 4 list specific η values used for different methods and datasets, e.g., 'SSW 10', 'SW 0.001', 'DSSW (exp) 10', 'DSSW-SSL (exp, η=100, L=200)'. |