Learning to Communicate with Strangers via Channel Randomisation Methods

Dylan Cope; Nandi Schoots

Learning to Communicate with Strangers via Channel Randomisation Methods

Dylan Cope, Nandi Schoots

Informatics

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

3 Downloads (Pure)

Abstract

We introduce two methods for improving the performance of agents meeting for the first time to accomplish a communicative task. The methods are: (1) `message mutation' during the generation of the communication protocol; and (2) random permutations of the communication channel. These proposals are tested using a simple two-player game involving a `teacher' who generates a communication protocol and sends a message, and a `student' who interprets the message. After training multiple agents via self-play we analyse the performance of these agents when they are matched with a stranger, i.e. their zero-shot communication performance. We find that both message mutation and channel permutation positively influence performance, and we discuss their effects.

Original language	English
Title of host publication	The 4th Workshop on Emergent Communication at the International Conference on Neural Information Processing (NeurIPS)
Publication status	Published - 2020

Keywords

Deep Learning
Reinforcement Learning
Emergent Communication
Domain Generalization
Out-Of-Distribution Generalisation
Multi-agent Learning
Multi-agent reinforcement learning

Access to Document

Learning to Communicate with Strangers (Camera Ready)Final published version, 1.12 MB

Cite this

@inbook{8b8cb9154bc94f7b93c975acb32479bd,

title = "Learning to Communicate with Strangers via Channel Randomisation Methods",

abstract = "We introduce two methods for improving the performance of agents meeting for the first time to accomplish a communicative task. The methods are: (1) `message mutation' during the generation of the communication protocol; and (2) random permutations of the communication channel. These proposals are tested using a simple two-player game involving a `teacher' who generates a communication protocol and sends a message, and a `student' who interprets the message. After training multiple agents via self-play we analyse the performance of these agents when they are matched with a stranger, i.e. their zero-shot communication performance. We find that both message mutation and channel permutation positively influence performance, and we discuss their effects.",

keywords = "Deep Learning, Reinforcement Learning, Emergent Communication, Domain Generalization, Out-Of-Distribution Generalisation, Multi-agent Learning, Multi-agent reinforcement learning",

author = "Dylan Cope and Nandi Schoots",

year = "2020",

language = "English",

booktitle = "The 4th Workshop on Emergent Communication at the International Conference on Neural Information Processing (NeurIPS)",

}

TY - CHAP

T1 - Learning to Communicate with Strangers via Channel Randomisation Methods

AU - Cope, Dylan

AU - Schoots, Nandi

PY - 2020

Y1 - 2020

N2 - We introduce two methods for improving the performance of agents meeting for the first time to accomplish a communicative task. The methods are: (1) `message mutation' during the generation of the communication protocol; and (2) random permutations of the communication channel. These proposals are tested using a simple two-player game involving a `teacher' who generates a communication protocol and sends a message, and a `student' who interprets the message. After training multiple agents via self-play we analyse the performance of these agents when they are matched with a stranger, i.e. their zero-shot communication performance. We find that both message mutation and channel permutation positively influence performance, and we discuss their effects.

AB - We introduce two methods for improving the performance of agents meeting for the first time to accomplish a communicative task. The methods are: (1) `message mutation' during the generation of the communication protocol; and (2) random permutations of the communication channel. These proposals are tested using a simple two-player game involving a `teacher' who generates a communication protocol and sends a message, and a `student' who interprets the message. After training multiple agents via self-play we analyse the performance of these agents when they are matched with a stranger, i.e. their zero-shot communication performance. We find that both message mutation and channel permutation positively influence performance, and we discuss their effects.

KW - Deep Learning

KW - Reinforcement Learning

KW - Emergent Communication

KW - Domain Generalization

KW - Out-Of-Distribution Generalisation

KW - Multi-agent Learning

KW - Multi-agent reinforcement learning

M3 - Conference paper

BT - The 4th Workshop on Emergent Communication at the International Conference on Neural Information Processing (NeurIPS)

ER -

Learning to Communicate with Strangers via Channel Randomisation Methods

Abstract

Keywords

Access to Document

Fingerprint

Cite this