next up previous
Next: Receiver Choice Functions Up: Using Decision Tree Confidence Previous: Layered Learning

Using the Learned Behaviors


As described in Section 2, ML techniques have been studied in the Soccer Server in isolated situations. However, the resulting behaviors have not yet been tested in full game situations. In this paper, we examine the effectiveness in game situations of the DT learned in [13].

To our knowledge, this paper reports the first use of confidence factors from DTs for agent control. In particular, the confidence factors that are returned along with classifications can be used to differentiate precisely among several options.

Peter Stone
Sun Dec 7 06:59:19 EST 1997