Towards a Better Global Loss Landscape of GANs

Authors: Ruoyu Sun, Tiantian Fang, Alexander Schwing

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic data show that the predicted bad basin can indeed appear in training. We also perform experiments to support our theory that Rp GAN has a better landscape than separable-GAN. For instance, we empirically show that Rp GAN performs better than separable-GAN with relatively narrow neural nets.
Researcher Affiliation Academia Ruoyu Sun , Tiantian Fang, Alex Schwing University of Illinois at Urbana-Champaign ruoyus,tf6,aschwing@illinois.edu
Pseudocode No The paper does not contain any blocks explicitly labeled 'Pseudocode' or 'Algorithm'.
Open Source Code Yes The code is available at https://github.com/Ailsa F/RS-GAN.
Open Datasets Yes For setting (A), we test on CIFAR-10 and STL-10 data.
Dataset Splits No The paper refers to using CIFAR-10 and STL-10, which have standard splits, but it does not explicitly state the train/validation/test percentages or sample counts for these datasets. It mentions evaluating generated samples (50k and 10k) but not the dataset splits themselves.
Hardware Specification No The paper does not specify any particular hardware used for experiments, such as GPU or CPU models.
Software Dependencies No The paper mentions using the Adam optimizer but does not specify software dependencies like programming language versions or library versions (e.g., PyTorch 1.x, TensorFlow 2.x).
Experiment Setup Yes For the optimizer, we use Adam with the discriminator s learning rate 0.0002. For CIFAR-10 on Res Net, we set β1 = 0 and β2 = 0.9 in Adam; for others, β1 = 0.5 and β2 = 0.999. We tune the generator s learning rate and run 100k iterations in total.