Systems | Information | Learning | Optimization

Game Redesign in No-regret Game Playing

Abstract: We all know how an adversary can hack supervised learning (the deepnets that can’t see pandas). Many know how an adversary can hack reinforcement learning (policy poisoning). Game playing is next. We study the game redesign problem in which an external designer has the ability to change the payoff function in each round, but incurs a design cost for deviating from the original game. The players apply no-regret learning algorithms to repeatedly play the changed games. The goals of the designer are to (i) incentivize all players to take a specific target joint action frequently; and (ii) incur small cumulative design cost. We present game redesign algorithms with the guarantee that the target action profile is played in T-o(T) rounds while incurring only o(T) cumulative design cost. Game redesign has positive purposes, too: a benevolent designer can incentivize players to take a target joint action with better social welfare compared to the solution of the original game.
December 1 @ 12:30
12:30 pm (1h)

Orchard View Room, Virtual

Jerry Zhu