On the learning behavior of adaptive networks — Part I: Transient analysis