Learning direction in global motion: two classes of psychophysically-motivated models