Structural Inductive Biases in Emergent Communication
Computation and Language
2021-07-28 v4 Artificial Intelligence
Machine Learning
Multiagent Systems
Machine Learning
Abstract
In order to communicate, humans flatten a complex representation of ideas and their attributes into a single word or a sentence. We investigate the impact of representation learning in artificial agents by developing graph referential games. We empirically show that agents parametrized by graph neural networks develop a more compositional language compared to bag-of-words and sequence models, which allows them to systematically generalize to new combinations of familiar features.
Cite
@article{arxiv.2002.01335,
title = {Structural Inductive Biases in Emergent Communication},
author = {Agnieszka Słowik and Abhinav Gupta and William L. Hamilton and Mateja Jamnik and Sean B. Holden and Christopher Pal},
journal= {arXiv preprint arXiv:2002.01335},
year = {2021}
}
Comments
The first two authors contributed equally. Poster presented at CogSci 2021