Emergence of Linguistic Communication from Referential Games with Symbolic and Pixel Input


Nov 03, 2017 (modified: Nov 03, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: Emergent communication problems can be used to study the ability of algorithms to evolve or learn communication protocols. In this work, we study the properties of protocols emerging when reinforcement learning agents are trained end-to-end on referential communication games. We extend previous work using symbolic representations to using raw pixel input data, a more challenging and realistic input representation. We find that the degree of structure found in the input data affects the nature of the emerged protocols, and thereby corroborate the hypothesis that structured language is most likely to emerge when agents perceive the world as being structured.
  • TL;DR: A controlled study of the role of environments with respect to properties in emergent communication protocols.
  • Keywords: disentanglement, communication, emergent language, compositionality, multi-agent