We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
1 parent 82e6d0c commit 5dd4444Copy full SHA for 5dd4444
dl/gan/gan.md
@@ -61,9 +61,8 @@ Discriminator是一个二元分类器,输入是图像,输出是两类:“
61
其次,要定义loss function才能训练。前面说了,GANs可以看成一个博弈,那么博弈双方都会有cost(代价),如果是零和博弈,那么双方的cost之和为0。Discriminator是一个分类器,它的loss可以定义用交叉熵来定义:
62
$$J^{(D)}(\theta^{(D)},\theta^{(G)})=-\frac {1}{2}E_{xP\sim _{data}}logD(x)-\frac {1}{2}E_zlog(1-D(G(z)))$$
63
如果是零和博弈,那么Generator的loss就定义为:
64
-$$J^{(D)}(\theta^{(D)},\theta^{(G)})=-J^{eta^{(D)},\theta^{(G)})=\frac {1}{2}E_{xP\sim _{data}}logD(x)+\frac {1}{2}E_zlog(1-D(G(z)))
65
-
66
+$$J^{(D)}(\theta^{(D)},\theta^{(G)})=-J^{eta^{(D)},\theta^{(G)})=\frac {1}{2}E_{xP\sim _{data}}logD(x)+\frac {1}{2}E_zlog(1-D(G(z)))$$
67
整个优化问题就是一个minmax博弈
68

69
0 commit comments