A Strategic Analysis of Multi-agent Protocols
Sieuwert van Otterloo
Abstract:
The research goal behind this dissertation is to develop ways to compare and analyse multi-agent protocols. In order to do so one has to distinguish different types of protocols, and one has to distinguish different classes of properties. Protocols that can be modelled as imperfect information game forms are therefore discussed in the first part of this dissertation, whereas protocols that can be modeled as imperfect information are the subject of the second part. In both parts we define concepts that help us to analyse and understand protocols, demonstrate these concepts on example protocols, and investigate the computational properties of these concepts.
In chapter 4, a logic for reasoning about what coalitions can achieve in protocols is presented. For this logic, a complete proof system is given, and the model checking complexity is determined.
In chapter 5, logics for reasoning about more complicated properties are presented. Specifically we compare the model checking complexity of logics for reasoning about side-effects and nested abilities.
In chapter 6, protocols are analysed using logics that deal with preferences explicitly. For two different variants of preference logics we give completeness proofs, and as an example, a characterisation of backward induction is given.
Protocols with imperfect information are the topic of the second part of this dissertation. In these protocols the knowledge that agents have plays a leading role. One can look at knowledge in a qualitative way, using epistemic logic, and this is done in chapter 7. In this chapter, it is shown how the computational complexity of protocol verification, depends on the presence of opponents, on whether strategies are known, and on the monotonic nature of the knowledge requirements. In chapter 8, it is shown that one can also model knowledge in a quantitative way. Using this approach, we compute optimal strategies for privacy preservation.