PSA2016: The 25th Biennial Meeting of the Philosophy of Science Association

Full Program »

Information dynamics in learned referential signalling games: two sources of entropy

Reinforcement learning in signalling games does not always converge on optimal systems, but some mechanisms allow it to do so. We look at a slightly different case, the \textit{referential} signalling game. This allows us to use an information theoretic approach not usually possible with signalling games, and to analyse optimal signalling in terms of two sources of entropy. This clarifies why certain mechanisms lead to signalling systems in referential games, which leads us to suggest a similar explanation for the non-referential game.

Author Information:

Matthew Spike    
Philosophy, College of Arts and Social Sciences
ANU

 

Powered by OpenConf®
Copyright©2002-2015 Zakon Group LLC