Parallel Algorithms Align with Neural Execution
Abstract
Neural algorithmic reasoners are parallel processors. Teaching them sequential algorithms contradicts this nature, rendering a significant share of their computations redundant. Parallel algorithms however may exploit their full computational power, therefore requiring fewer layers to be executed. This drastically reduces training times, as we observe when comparing parallel implementations of searching, sorting and finding strongly connected components to their sequential counterparts on the CLRS framework. Additionally, parallel versions achieve strongly superior predictive performance in most cases.
Cite
Text
Engelmayer et al. "Parallel Algorithms Align with Neural Execution." Proceedings of the Second Learning on Graphs Conference, 2023.Markdown
[Engelmayer et al. "Parallel Algorithms Align with Neural Execution." Proceedings of the Second Learning on Graphs Conference, 2023.](https://mlanthology.org/log/2023/engelmayer2023log-parallel/)BibTeX
@inproceedings{engelmayer2023log-parallel,
title = {{Parallel Algorithms Align with Neural Execution}},
author = {Engelmayer, Valerie and Georgiev, Dobrik Georgiev and Veličković, Petar},
booktitle = {Proceedings of the Second Learning on Graphs Conference},
year = {2023},
pages = {31:1-31:13},
volume = {231},
url = {https://mlanthology.org/log/2023/engelmayer2023log-parallel/}
}