An Interactive Search Game with Two Agents

Xing Gao, Lev Reyzin

Abstract: We study a two-player interactive game where the players search for a target vertex in a connected undirected graph via multiple rounds of query and feedback. This two-player search game is an adaptation of the single-agent interactive learning problem. Here, we analyze the game for trees under different costs: under zero-sum cost we present the one-step Nash equilibrium strategy, and show that the competition hinders both players’ search progress as compared to binary search; under non-zero-sum costs, however, we present a cooperative strategy which benefits both players.