Finding Mixed Nash Equilibria of Generative Adversarial Networks
Authors: Ya-Ping Hsieh, Chen Liu, Volkan Cevher
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we provide experimental evidence that our approach yields comparable or superior results to contemporary training algorithms, and outperforms classical methods such as SGD, Adam, and RMSProp. |
| Researcher Affiliation | Academia | Ya-Ping Hsieh 1 Chen Liu 1 Volkan Cevher 1 1LIONS, EPFL, Switzerland. |
| Pseudocode | Yes | Algorithm 1 INFINITE-DIMENSIONAL ENTROPIC MD; Algorithm 2 INFINITE-DIMENSIONAL ENTROPIC MP; Algorithm 3 MIRROR-GAN: APPROXIMATE MIRROR DECENT FOR GANS |
| Open Source Code | No | The paper does not explicitly state that source code for the described methodology is available, nor does it provide a link to a code repository. |
| Open Datasets | Yes | We repeat the synthetic setup as in (Gulrajani et al., 2017). The tasks include learning the distribution of 8 Gaussian mixtures, 25 Gaussian mixtures, and the Swiss Roll. ... For real images, we use the LSUN bedroom dataset (Yu et al., 2015). ... We have also conducted a similar study with MNIST |
| Dataset Splits | No | The paper mentions using datasets for experiments but does not provide specific details on how the data was split into training, validation, or test sets (e.g., percentages or sample counts), nor does it reference standard predefined splits for these datasets. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU or CPU models, or cloud computing instance types used for running the experiments. |
| Software Dependencies | No | The paper mentions several algorithms and architectures like SGD, Adam, RMSProp, and DCGAN, but does not specify software dependencies with version numbers (e.g., Python version, specific library versions like PyTorch or TensorFlow). |
| Experiment Setup | Yes | For both the generator and discriminator, we use two MLPs with three hidden layers of 512 neurons. We use the same architecture (DCGAN) as in (Radford et al., 2015) with batch normalization. The step-sizes for all algorithms are determined via parameter sweeping. (Also confirmed details in Appendix D, e.g., 'The initial learning rate is 10^-4 for all cases and for all algorithms. ... For Mirror-GAN and Mirror-Prox-GAN, the parameters for the SGLD are γ = 10−4, β = 0.5, and Kt = 100.') |