Application of Learning to Trust-Adaptive Agents

In this paper we analyse and evaluate, in which ways learning techniques can be applied to agents in an open system, which have to map continuous situations into a continuous action space. The agents are part of an open desktop grid, where agents can offer and use computational power of other volunteer agents in order to improve their speedup for bag of-task applications. Moreover, the agents use a trust-based mechanism, which enables the system to exclude misbehaving agents from the community. In this paper, the decision mechanism of such agents is enhanced using learning techniques to determine optimal cooperation thresholds.