In an open trusted Desktop Grid system with a normative environment incentives and sanctions may change during runtime. Every agent in the system computes work for other agents and also submits jobs to other agents. It has to decide for which agents it wants to work and to which agent it wants to give its jobs. We introduced a trust metric to isolate misbehaving agents. After getting a job processed by another agent it will get a reward. When processing a job for another agent it will get a positive trust-rating, but no direct reward. To come to a decision when accepting or rejecting jobs we need to be able to estimate the reward. Since the environment may change at runtime and to overcome delayed reward issues we use a neural network to estimate the reward based on the environment and trust level.
展开▼
