Agents in an open and distributed environment can achieve their goals better and faster if they can pre-dict future states of their environment. Previous works on predictive agents showed that very simple predic-tion strategies alone are not sufficient in order to deal with complex dynamics (Kephart, Hogg & Huberman 1990) and often took the way of developing a very spe-cialized model in order to predict a highly dynamic model-environment (Hubler &c Pines 1993). Regarding this tradeoff between accuracy and speed we argue that there might be situations where agents have to react in a real-time manner and are doing better by preferring speed over accuracy. Our agents applying Linear Pre-diction observe their environment as a series of time discrete events. This environment could be e.g. the number of agents using a resource at a certain time.
展开▼