关于Wasserstein Distance的计算,似乎还有一个简单一点的推导方法,在《BEGAN: Boundary Equilibrium Generative Adversarial Networks》中给出了一个推导的过程:
Wasserstein Distance的定义式如下:
W(ℙr,ℙg)=infγ∈Π(ℙr,ℙg)E(x,y)∼γ[‖x−y‖](1) W ( P r , P g ) = inf γ ∈ Π ( P r , P g ) E ( x , y ) ∼ γ [ ‖ x − y ‖ ] ( 1 )
根据Jensen不等式有:
infE[|x1−x2|]≥inf|E[x1−x2]|=|E(x1)−E(x2)|(2) inf E [ | x 1 − x 2 | ] ≥ inf | E [ x 1 − x 2 ] | = | E ( x 1 ) − E ( x 2 ) | ( 2 )
于是代入(1)有:
W(ℙr,ℙg)≥|Ex∼ℙr(fθ(x))−Ex∼ℙg(fθ(x))|=|