Erdos-Renyi:
1
?
(
n
l
?
1
+
n
l
)
/
(
n
l
?
1
?
n
l
)
1-(n^{l-1}+n^l)/(n^{l-1}*n^l)
1?(nl?1+nl)/(nl?1?nl)
Erdos-Renyi-Kernel(ERK):
1
?
(
n
l
?
1
+
n
l
+
w
l
+
h
l
)
/
(
n
l
?
1
?
n
l
?
w
l
?
h
l
)
1-(n^{l-1}+n^l+w^l+h^l)/(n^{l-1}*n^l*w^l*h^l)
1?(nl?1+nl+wl+hl)/(nl?1?nl?wl?hl)
每层的参数量越大,可以分配更大的sparsity,ERK的实验效果最好
[WACV 2022] Hessian-Aware Pruning and Optimal Neural Implant