Do GANs always have Nash equilibria?
Authors: Farzan Farnia, Asuman Ozdaglar
ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we show through theoretical and numerical results that indeed GAN zero-sum games may have no Nash equilibria. We perform several numerical experiments indicating the existence of proximal equilibria in GANs. We empirically examined whether standard GAN architectures converge to Nash equilibrium solutions. In our numerical experiments, we applied three standard GAN architectures... to standard MNIST and Celeb A datasets. |
| Researcher Affiliation | Academia | 1Laboratory for Information & Decision Systems, Massachusetts Institute of Technology, Cambridge, MA, USA. |
| Pseudocode | Yes | Algorithm 1 GAN Proximal Training |
| Open Source Code | No | The paper mentions using existing open-source implementations for baseline GAN architectures (e.g., "using the (Gulrajani et al., 2017) s implementation of Wasserstein GANs with the code available at the paper s Github repository" and "we used the implementations of (Miyato et al., 2018; Farnia et al., 2019)"), but it does not provide explicit access (link, statement) to its own code for the proposed proximal training methodology. |
| Open Datasets | Yes | In our numerical experiments, we applied three standard GAN architectures... to standard MNIST (Le Cun, 1998) and Celeb A (Liu et al., 2015) datasets. Applying the proximal training to MNIST, CIFAR-10, and Celeb A datasets |
| Dataset Splits | No | The paper mentions using standard datasets like MNIST, Celeb A, and CIFAR-10, and describes training iterations and discriminator updates. However, it does not explicitly provide specific details on how the data was split into training, validation, and test sets (e.g., percentages, sample counts, or explicit standard split names for all datasets). |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU models, CPU types, or memory specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using specific optimizers like Adam and RMSprop and states that they used existing implementations of GANs (e.g., Gulrajani et al., 2017; Miyato et al., 2018), but it does not provide specific version numbers for general software dependencies or libraries (e.g., Python version, specific deep learning framework version like TensorFlow or PyTorch). |
| Experiment Setup | Yes | In the experiments, we used the DCGAN 4-layer CNN architecture for both the discriminator and generator functions... ran each experiment for 200,000 generator iterations with 5 discriminator updates per generator update. We used the RMSprop optimzier... or the Adam optimizer... We computed the gradient of the above proximal objective by applying the Adam optimizer for 50 steps to approximate the solution to the proximal optimization... |