Noise Detection in Agent Reputation Models Using IMM Filtering

Abstract

Inferring Trust in dynamic and subjective environments is a key interesting issue in the way to obtain a complete delegation of human-like decisions in autonomous agents. With this final intention several trust models and strategies have been proposed by researchers, and some of them were tested using the Agent Reputation and Trust (ART) testbed. competitions. In this paper we propose to apply a temporal statistical model to the noisy observations perceived (reputations) that filter out noise (subjectivity) and estimate future state variable (trust). Specifically we have implemented agents that apply Kalman and Interacting Multiple Model (IMM) adaptive filters as a part of ART trust models. Kalman and IMM have been largely applied to make time-dependent predictions in noisy environments, furthermore they are recognized as a reasoning paradigm for time-variable facts, so they seem to be appropriate for agents to infer trust from direct observations and indirect references. In order to show its viability as part of a trust model, we have run ART games with other agents that took part in past ART competitions.