Just return to this post to mention a rather obvious assumption that following the norms assumes that these norms are stable throughout all the time and space duration of the group, which is clearly not true for large, old or internally diverse groups. Thus, some model of the group boundaries and-or internal structure should be either hand-coded, or meta-learned before norm-learning.
Just return to this post to mention a rather obvious assumption that following the norms assumes that these norms are stable throughout all the time and space duration of the group, which is clearly not true for large, old or internally diverse groups. Thus, some model of the group boundaries and-or internal structure should be either hand-coded, or meta-learned before norm-learning.
Yes, I agree that you need to deal with the fact that norms are not stable across time and space.