Storchastic: A Framework for General Stochastic Automatic Differentiation

Authors: Emile Krieken, Jakub Tomczak, Annette Ten Teije

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we implement Storchastic as a Py Torch library at github.com/HEmile/storchastic. ... We run this model on our currently implemented set of gradient estimation methods for discrete variables in Appendix E and report the results, which are meant purely to illustrate the case study.
Researcher Affiliation Academia Emile van Krieken Vrije Universiteit Amsterdam e.van.krieken@vu.nl Jakub M. Tomczak Vrije Universiteit Amsterdam j.m.tomczak@vu.nl Annette ten Teije Vrije Universiteit Amsterdam annette.ten.teije@vu.nl
Pseudocode Yes Algorithm 1 The Storchastic framework: Compute a Monte Carlo estimate of the n-th order gradient given k gradient estimators qi, wi, li, ai .
Open Source Code Yes Finally, we implement Storchastic as a Py Torch library at github.com/HEmile/storchastic. ... We implemented Storchastic as an open source Py Torch [39] library 1. 1Code is available at github.com/HEmile/storchastic.
Open Datasets Yes 2.3 Example: Discrete Variational Autoencoder ... We run this model on our currently implemented set of gradient estimation methods for discrete variables in Appendix E and report the results, which are meant purely to illustrate the case study. ... For our experiments, we use the binarized MNIST dataset.
Dataset Splits No The paper describes using a dataset (MNIST) but does not provide specific details on how it was split into training, validation, and test sets. Appendix E mentions 'We train all models for 100 epochs with Adam [24] and a learning rate of 1e-3' but no explicit splits.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, or cloud instance specifications).
Software Dependencies No The paper mentions 'Py Torch' but does not specify a version number. No other software dependencies with specific version numbers are listed.
Experiment Setup Yes In Figure 4, we show how to implement the discrete VAE. The implementation directly follows the SCG shown in Figure 2. ... In Appendix E, it states: 'We train all models for 100 epochs with Adam [24] and a learning rate of 1e-3'.