Exploring the performance of a baseline for value based statistical algorithms on dynamic Iterated Prisoner’s Dilemma networks

Abstract

We conducted research on inspecting the performance of a value-based statistical algorithm on a network where each agent pair plays an extension of the iterated prisoner’s dilemma game (IPD), which supports choosing and refusing partners, also known as IPD/CR. Here we show by multi-agent simulation that this simple statistical algorithm has comparable performance to other well-known strategies on IPD games such as tit-for-tat (TFT) and win-stay-lose-shift (Pavlov). We also add noise corruption to the network to simulate networks in real scenarios where agents may misunderstand the passed message, and we compare the robustness of several well-known IPD strategies as well as the proposed statistical algorithm