Using adaptive consultation of experts to improve convergence rates in multiagent learning
Speaker: Greg Hines
In this paper we study the use of experts algorithms in a multiagent setting. In this paper we allow agents to use multiple experts and explore different experts algorithms that allow agents to adaptively decide which expert to consult. By doing so, agents can further improve the learning process and are better able to deal with new games and situations.
Reducing communicating cost: A mechanism design approach
Speaker: Michelle Zhang
In this paper we study a problem where self-interested agents must interact with some centralized mechanism where this interaction is costly. Instead, agents may choose to interact with neighbours in order to coordinate their actions, potentially resulting in savings with respect to total interaction costs for all involved. We highlight the issues that arise in such a setting for the mechanism as well as for the agents. We propose a cost-sharing mechanism that agents can use to coordinate and reduce their interaction costs. We also discuss how agents might form groups. Experimental works validate our model and proposals.