Structural Inductive Biases in Emergent Communication
arXiv:2002.01335
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.
arXiv commentThe first two authors contributed equally. Poster presented at CogSci 2021
References in corpus (12)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Sequence to Sequence Learning with Neural Networks
- Inductive Representation Learning on Large Graphs
- Categorical Reparameterization with Gumbel-Softmax
- The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
- Learning Multiagent Communication with Backpropagation
- Emergence of Grounded Compositional Language in Multi-Agent Populations
- On the interaction between supervision and self-play in emergent communication
- Measuring Compositionality in Representation Learning
- How agents see things: On visual representations in an emergent language game
- Inductive Bias and Language Expressivity in Emergent Communication
- Towards Graph Representation Learning in Emergent Communication