Official repository for the paper "Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks"
MIT License
(February 2022) This repository is no longer being maintained. All the experiments can be done with new and improved code. See our new deep-thinking repository.
The official repository for the paper "Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks".
To install requirements:
pip install -r requirements.txt
To use the datasets we use in this project, we recommend you install our Python package easy-to-hard-data
by running:
pip install easy-to-hard-data
You many also download raw datasets. See the Google Drive folder.
See the dataset specific documentation in the corresponding directories: Prefix Sums, Mazes, Chess.
If you find this code helpful, please consider citing our work.
@misc{schwarzschild2021learn,
title={Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks},
author={Avi Schwarzschild and Eitan Borgnia and Arjun Gupta and Furong Huang and Uzi Vishkin and Micah Goldblum and Tom Goldstein},
year={2021},
eprint={2106.04537},
archivePrefix={arXiv},
primaryClass={cs.LG}
}