A-NICE-MC: Adversarial Training for MCMC

Authors: Jiaming Song, Shengjia Zhao, Stefano Ermon

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

Reproducibility Variable Result LLM Response
Research Type Experimental Empirical results demonstrate that A-NICE-MC combines the strong guarantees of MCMC with the expressiveness of deep neural networks, and is able to significantly outperform competing methods such as Hamiltonian Monte Carlo.
Researcher Affiliation Academia Jiaming Song Stanford University tsong@cs.stanford.edu Shengjia Zhao Stanford University zhaosj12@cs.stanford.edu Stefano Ermon Stanford University ermon@cs.stanford.edu
Pseudocode No No structured pseudocode or algorithm blocks were found in the paper.
Open Source Code Yes Code for reproducing the experiments is available at https://github.com/ermongroup/a-nice-mc.
Open Datasets Yes We experiment with a distribution pd over images, such as digits (MNIST) and faces (Celeb A).
Dataset Splits No The paper mentions training and testing, but does not explicitly provide details for validation dataset splits or specific use of a validation set for hyperparameter tuning.
Hardware Specification No The paper states 'All training and sampling are done on a single CPU thread' but does not specify the CPU model, type, or other hardware details.
Software Dependencies No The paper mentions 'TensorFlow [38]' and 'Adam [39] as optimizer' but does not provide specific version numbers for these software components.
Experiment Setup Yes All the models are trained with the gradient penalty objective for Wasserstein GANs [17, 15], where λ = 1/3, B = 4 and M = 3.