发帖
雷达卡
Elementary Superexpressive Activations
Dmitry Yarotsky 1
Abstract (i.e., for d = 1). The key idea here is to use the separability
We call a finite family of activation functions su- of the space C([0, 1]), and construct a (quite complicated)
perexpressive if any multivariate continuous func- activation by joining all the functions from some dense
tion can be approximated by a neural network that countable subset. In ...
京公网安备 11010802022788号
