Abstract
Because of the complexity of multi-agent systems,
machine learning techniques are indispensable.
When the agents cooperate with each other,
they have to make agreement about next behaviors.
But, if some good cooperative strategies have almost
the same effectiveness, agreement could be unstable,
and efficiency of learning will reduce.
In this paper, we show that agent can avoid this problem
with incremental learning.
We evaluate the effectiveness using a case study of
learning pass play on the Soccer Server.