在機器學習裡,經常會看到兩個詞,data fidelity term, regularization(prior) term.
例如,在image restoration中,我們需要最小化如下的object function(MAP, 最大後驗機率),則:
data fidelity term:
regularization(prior) term:
那麼兩項的作用是什麼呢?
The fidelity term guarantees the solution accords with the degradation process,
翻譯:資料保真項保證結果符合降質過程
while the regularization term enforces desired property of the output.
翻譯:正則(先驗)項對輸出進行增強
Since IR is an ill-posed inverse problem, the prior which is also called regularization needs to be adopted to constraint the solution space
翻譯:因為image restoration是病态解問題,是以用正則(先驗)項限制解空間