In many systems, agents must rely on their peers to achieve their goals. However, when trusted to perform an action, an agent may betray that trust by not behaving as required. Agents must therefore estimate the behaviour of their peers, so that they may identify reliable interaction partners. To this end, we present a Bayesian trust model (HABIT) for assessing trust based on direct experience and (potentially unreliable) reputation. Although existing approaches claim to achieve this, most rely on heuristics with little theoretical foundation. In contrast, HABIT is based on principled statistical techniques; can be used with any representation of behaviour; and can assess trust based on observed similarities between groups of agents. In this paper, we describe the theoretical aspects of the model and present experimental results in which HABIT was shown to be up to twice as accurate at predicting trustee performance as an existing state-of-the-art trust model.
Dr WT Luke Teacy
Luke received his PhD from the University of Southampton on the topic of Trust and Reputation in Multi-Agent Systems. He was subsequently a post-doc on the project Adaptive Energy Aware Sensor Networks, also at Southampton. He recently moved to the University of Ulster, where he'll be working on Autonomous Control of Swarms of Unmanned Aerial Vehicles.
His research interests include trust and reputation systems, decentralised control and learning in multi-agent systems.