Approximate Optimal Transport for Continuous Densities with Copulas
Authors: Jinjin Chi, Jihong Ouyang, Ximing Li, Yang Wang, Meng Wang
IJCAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Empirical results on real applications of image retrieval and synthetic data demonstrate that our Cop-OT can gain more accurate approximations to continuous OT values than the state-of-the-art baselines. ... In this section, we empirically evaluate Cop-OT on both synthetic and real data. |
| Researcher Affiliation | Academia | Jinjin Chi1,2 , Jihong Ouyang1,2 , Ximing Li1,2 , Yang Wang3 and Meng Wang3 1 College of Computer Science and Technology, Jilin University, China 2 Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education, China 3 School of Computer Science and Information Engineering, Hefei University of Technology, China |
| Pseudocode | Yes | Algorithm 1 Optimization of Cop-OT |
| Open Source Code | Yes | The derivation details of Eq.11 can be found at https://github.com/jinjinchi/Approximate-Optimal-Transport-for-Continuous-Densities-with-Copulas. |
| Open Datasets | Yes | The MINST5, a dataset of handwritten digits from zero to nine, is used. We randomly select 10,000 images as the database and 150 images for texting. ... 5http://yann.lecun.com/exdb/mnist/ |
| Dataset Splits | No | The paper states using 10,000 images as a database and 150 for testing, but does not specify training, validation, or test splits beyond that. It does not mention cross-validation or specific percentages for these splits. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used, such as GPU/CPU models, memory, or cloud instances. It only mentions the "expensive computational cost on neural networks" in a general sense. |
| Software Dependencies | No | The paper mentions the Adam method [Kingma and Ba, 2015] and Vine Copula, but does not provide specific version numbers for these or other software dependencies. |
| Experiment Setup | Yes | For Cop-OT, we employ the family of Gaussian copula function, and use the standard Gaussians as the mapping distribution in the reparameterization trick. For all methods. the sample number S is set to 200, and we report the average results of five independent runs. ... We use the Adam method [Kingma and Ba, 2015] to adaptively adjust the optimization process... in this work we empirically fixed the parameters of Adam as follows: β1 = 0.9, β2 = 0.999 and α = 0.001. |