This paper presents a framework for understanding the emergence and analysis of language in social learning agents using reinforcement learning in grid-world mazes. A teacher network transmits a message to a student network to improve task performance. The framework compresses high-dimensional task information into a low-dimensional representational space, mimicking natural language features. Results show that teacher information improves task completion and generalization, and optimizing message content enhances information encoding, highlighting language's role as a common representation among agents.
Publisher
Nature Communications
Published On
Aug 31, 2024
Authors
Tobias J. Wieczorek, Tatjana Tchumatchenko, Carlos Wert-Carvajal, Maximilian F. Eggl
Tags
social learning agents
language emergence
reinforcement learning
task performance
grid-world mazes
information encoding
teacher-student network
Related Publications
Explore these studies to deepen your understanding of the subject.
