An Online Learning Approach to Generative Adversarial Networks
Authors: Paulina Grnarova, Kfir Y Levy, Aurelien Lucchi, Thomas Hofmann, Andreas Krause
ICLR 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | On several real-world tasks our approach exhibits improved stability and performance compared to standard GAN training. |
| Researcher Affiliation | Academia | Paulina Grnarova, Kfir Y. Levy, Aurelien Lucchi, Thomas Hofmann, Andreas Krause ETH Zürich {paulina.grnarova,yehuda.levy,aurelien.lucchi,thomas.hofmann}@inf.ethz.ch krausea@ethz.ch |
| Pseudocode | Yes | Algorithm 1 CHEKHOV GAN, Algorithm 2 Practical CHEKHOV GAN, Algorithm 3 Update queue for Algorithm A1 and A2 |
| Open Source Code | No | The paper does not mention providing open-source code for the methodology. |
| Open Datasets | Yes | We train a simplified DCGAN architecture (see details in Appendix D) with both GAN and CHEKHOV GAN with a different number of saved past states... augmented version of the MNIST dataset...We turn to the evaluation of our model for the task of generating rich image data... train a DCGAN architecture on CIFAR10 (Krizhevsky & Hinton, 2009)... We estimate the number of missing modes on the Celeb A dataset (Liu et al., 2015) |
| Dataset Splits | Yes | We train for 30 epochs, which we find to be the optimal number of training steps for vanilla GAN in terms of MSE on images from the validation set. |
| Hardware Specification | No | No specific hardware details (e.g., CPU, GPU models, or cloud instances) are provided. |
| Software Dependencies | No | The paper mentions optimizers like 'RMSProp optimizer' and 'Adam Kingma & Ba (2014)' but does not provide version numbers for these or other software libraries. |
| Experiment Setup | Yes | The optimal learning rate for GAN is 0.001, and for CHEKHOV GAN is 0.01. For all CHEKHOV GAN models we use regularization of 0.1 for the discriminator and 0.0001 for the generator. The regularization is L2 regularization only on the fully connected layers... The learning rate for all the models is 0.0002 for both the generator and the discriminator and the updates are performed using the Adam optimizer. |