View Single Post
Old 08-17-2012, 09:12 PM
yaser's Avatar
yaser yaser is offline
Join Date: Aug 2009
Location: Pasadena, California, USA
Posts: 1,478
Default Re: backpropagation at the final layer

Originally Posted by jianpan View Post
since \delta_1^{(L)}=\frac{\partial e(w)}{\partial s_1^{(L)}}, and e(w)=(x_1^{(L)}-y_n)^2, and x_1^{(L)}=\theta(s_1^{(L)}), I got \delta_1^{(L)}=2*(x_1^{(L)}-y_n)*(1-\theta^2(s)). is this correct? I found a python program at wikipedia about backpropagation using the \delta_1^{(L)}=(x_1^{(L)}-y_n)*(1-\theta^2(s)), and couldn't figure out where the 2 was dropped. Can someone confirm my derivation is correct? Thanks.
Some people define the error with a factor of 1 \over 2 in it in anticipation of the differentiation, hence the discrepancy.
Where everyone thinks alike, no one thinks very much
Reply With Quote