Generalization and Equilibrium in Generative Adversarial Nets (GANs)

Authors: Sanjeev Arora, Rong Ge, Yingyu Liang, Tengyu Ma, Yi Zhang

ICML 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we first explore the qualitative benefits of our method on image generation tasks: MNIST dataset (Le Cun et al., 1998) of hand-written digits and the Celeb A (Liu et al., 2015) dataset of human faces. Then for more quantitative evaluation we use the CIFAR-10 dataset (Krizhevsky & Hinton, 2009) and use the Inception Score introduced in (Salimans et al., 2016).
Researcher Affiliation Academia 1Princeton University, Princeton NJ 2Duke University, Durham NC.
Pseudocode No The paper describes the MIX+GAN protocol but does not provide structured pseudocode or an algorithm block.
Open Source Code Yes Related code is public online at https://github.com/ Princeton ML/MIX-plus-GANs.git
Open Datasets Yes MNIST dataset (Le Cun et al., 1998) of hand-written digits and the Celeb A (Liu et al., 2015) dataset of human faces. Then for more quantitative evaluation we use the CIFAR-10 dataset (Krizhevsky & Hinton, 2009)
Dataset Splits No The paper mentions training on datasets but does not explicitly provide training, validation, or test dataset splits (e.g., percentages or counts).
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper mentions software components and techniques (e.g., ADAM, DCGAN, WASSERSTEINGAN) but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We use exponentiated gradient (Kivinen & Warmuth, 1997): store the log-probabilities { ui, i 2 [T]}, and then obtain the weights by applying soft-max function on them: wui = e ui PT k=1 e uk , i 2 [T]. ... with learning rate lr = 0.0001. ... mixtures of 5 generators and 5 discriminators are used.