Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
DisARM: An Antithetic Gradient Estimator for Binary Latent Variables
Authors: Zhe Dong, Andriy Mnih, George Tucker
NeurIPS 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We evaluate Dis ARM on several generative modeling benchmarks and show that it consistently outperforms ARM and a strong independent sample baseline in terms of both variance and log-likelihood. |
| Researcher Affiliation | Industry | Zhe Dong Google Research, Brain Team EMAIL Andriy Mnih Deep Mind EMAIL George Tucker Google Research, Brain Team EMAIL |
| Pseudocode | Yes | we followed the techniques used in (Tucker et al., 2017; Grathwohl et al., 2018; Yin and Zhou, 2019) to extend Dis ARM to this setting (summarized in Appendix Algorithm 1). |
| Open Source Code | Yes | Code and additional information: https://sites.google.com/view/disarm-estimator. |
| Open Datasets | Yes | We evaluate the gradient estimators on three benchmark generative modeling datasets: MNIST, Fashion MNIST and Omniglot. |
| Dataset Splits | Yes | We use the standard split into train, validation, and test sets. |
| Hardware Specification | Yes | Furthermore, training the model on a P100 GPU was nearly twice as slow for RELAX, while ARM, Dis ARM and REINFORCE LOO trained at the same speed. |
| Software Dependencies | No | The paper mentions using "Adam (Kingma and Ba, 2015)" as an optimizer but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | The models were trained with Adam (Kingma and Ba, 2015) using a learning rate 10 4 on mini-batches of 50 examples for 106 steps. |