* ?0 {3 K# g8 S' m2 well, I can reason a bit from the names of these optimization types. The advantage of randomized optimization is for looking for global minimum without being trapped by a local minimum (which is often the case for deterministic optimization). so far (a few years ago before I left school, to be accurate) optimization is cursed by dimensionality, and random optimization has only limited success. 3 `4 T d( R2 R* b5 c5 N; U
/ r# }5 |$ C' _: t1 W3 ^
3. 没听说过傅里叶空间插值。如果有的话也不奇怪。那么像实空间插值类似,傅里叶空间插值能准确复原已知频率的结果。别的还有什么呢? : k/ \8 b# K9 V3 S( r* ?. S. I( m& [ ( g8 z M$ p0 n0 |4.不知道。我只知道复数比较奇妙。有个柯西定理,复函数如果一阶可导,则无穷阶可导。这在实函数是不可能的。* k# A% ^7 {& J4 e- q
可是本质区别是什么哪? - ]! M4 N8 y6 ]* \$ \" A/ o3 w$ Y7 h
5.一样大。/ c# z. k* `7 ], D/ h
) a2 }$ r' j; ?+ ~9 ~0 N2 v
6.蒙特卡洛的实质?我也不知道。它的误差是 O(1/sqrt(N)), N是sample个数。3 ?* H6 e/ y) H) I- G' Y
6 R9 U/ c3 |5 ?( l/ r# I* r! C