CS 285: Lecture 2, Imitation Learning. Part 4

[@user-ro7id8my9x]

Hi, I' wondering whether the second distributional shift is that the targets of training and testing are different?

[@snl92]

Only distributions introduced here with multitask are conditional (over task) distributions (so far as I can see), so these might be causing distributional shift. Not sure about different targets, I don't think so.