Fitting an inference algorithm instead of a model
Basically, forget about the distribution that the CRF represents, and instead only care how accurate are the results that pop out of inference.
MRF/CRF 的 inference 一般是用 MAP,这样会比较慢。文中的思路是放弃使用 MAP,放弃 CRF 代表的分布含义(ML学习用的就是CRF的分布)。转而采用高速但非最优的 inference algorithm A (譬如 LBP, Graph cut 等许多循环算法只采用4次循环), 通过最小化 x 和 t (true value) 之间的 loss 来学习参数。
Subscribe to:
Post Comments (Atom)
0 comments:
Post a Comment