Hybrid learning in signalling games

Lewis-Skyrms signaling games (Lewis 1969; Skyrms 2010) have been studied under a variety of low-rationality learning dynamics (Barrett 2006; Barrett and Zollman 2009; Huttegger, Skyrms, Smead, and Zollman 2010; Huttegger, Skyrms, Tarrès, and Wagner 2014; Huttegger, Skyrms, and Zollman 2014). Reinforcement dynamics are stable but slow and prone to evolving suboptimal signaling conventions. A low-inertia trial-and-error dynamical like win-stay/lose-randomize is fast and reliable at finding perfect signaling conventions but unstable in the context of noise or agent error. Here we consider a low-rationality hybrid of reinforcement and win-stay/lose-randomize learning that exhibits the virtues of both. This hybrid dynamics is reliable, stable, and exceptionally fast.