Gradient descent GAN optimization is locally stable

Authors: Vaishnavh Nagarajan, J. Zico Kolter

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

Reproducibility Variable Result LLM Response
Research Type Experimental In practice, the approach is simple to implement, and preliminary experiments show that it helps avert mode collapse and leads to faster convergence. We very briefly present experimental results that demonstrate that our regularization term also has substantial practical promise.
Researcher Affiliation Academia Vaishnavh Nagarajan Computer Science Department Carnegie-Mellon University Pittsburgh, PA 15213 vaishnavh@cs.cmu.edu J. Zico Kolter Computer Science Department Carnegie-Mellon University Pittsburgh, PA 15213 zkolter@cs.cmu.edu
Pseudocode No The paper describes algorithms and updates in text and mathematical equations, but it does not include any explicitly labeled "Pseudocode" or "Algorithm" blocks.
Open Source Code Yes We provide an implementation of this technique at https://github.com/locuslab/gradient_ regularized_gan
Open Datasets Yes We compare our gradient regularization to 10-unrolled GANs on the same architecture and dataset (a mixture of eight Gaussians) as in Metz et al. [2017]. We see similar results... in the case of a stacked MNIST dataset using a DCGAN [Radford et al., 2016]
Dataset Splits No The paper mentions datasets like "a mixture of eight Gaussians" and "stacked MNIST dataset", but it does not provide specific details on how these datasets were split into training, validation, and testing sets (e.g., percentages, sample counts, or explicit splitting methodology).
Hardware Specification No The paper does not provide any specific hardware details such as GPU models, CPU types, or memory configurations used for running the experiments.
Software Dependencies No The paper mentions the use of "modern automatic differentiation tools" but does not specify any software dependencies with version numbers (e.g., specific deep learning frameworks like PyTorch or TensorFlow, or their versions).
Experiment Setup No While the paper describes its proposed regularization term and mentions a specific regularization coefficient (e.g., "= 0.5" in Figure 1), it does not provide a comprehensive set of hyperparameters (such as learning rates, batch sizes, optimizer choices) or detailed training configurations required to fully reproduce the experimental setup.