[3L3-GS-8] Robot and real worlds: symbol emergence / knowledge extraction

[3L3-GS-8-03] Multiagent Inter-GMM: Extension and validation of symbol emergence model to more than three agents

〇Jun Inukai¹, Yoshinobu Hagiwara¹, Akira Taniguti¹, Tadahiro Taniguti¹ (1. Ritsumeikan University Emergent Systems Laboratory )

Keywords:Symbol emergence in robotics, Multi-agent

A system that builds a bottom-up symbolic system used by a group of agents for communication and cognition is called a symbol emergence system. The computational model of the symbol emergence system proposed by Hagiwara et al. is challenging to extend to three parties. The model is based on Metropolis-Hastings algorithm-based naming game. This study proposed three-agent version of the Metropolis-Hastings algorithm-based naming game. The naming game is proposed based on the modified version of the probabilistic model used in the prior study. The key idea of our proposal is based on a fact that the naming game between two agent can be regarded as a Markov chain Monte Carlo-based sampling procedure. We conducted experiments using synthetic data. The results confirmed that the proposed three-agent naming game worked appropriately, and the agent could successfully form categories and share signs. The adjusted Rand index and Kappa coefficients converged as the iteration progressed.