Paper: https://arxiv.org/abs/2007.10412
This repository contains code for all three types of experiments used in the paper:
- Feedforward network experiments described in
nn_experiments/README_ff.md - Recurrent network experiments described in
nn_experiments/README_rnn.md - PDE experiments described in
diffusionsim_experiments/README_diffusionsim.md
For the neural network experiments, the layers using randomized autodiff are in nn_experiments/layers.py.
To cite this work, please use
@misc{oktay2020randomized,
title={Randomized Automatic Differentiation},
author={Deniz Oktay and Nick McGreivy and Joshua Aduol and Alex Beatson and Ryan P. Adams},
year={2020},
eprint={2007.10412},
archivePrefix={arXiv},
primaryClass={cs.LG}
}