Skip to main content
12 events
when toggle format what by license comment
Nov 22, 2018 at 19:03 history bumped CommunityBot This question has answers that may be good or bad; the system has marked it active so that they can be reviewed.
Oct 15, 2018 at 3:00 history bumped CommunityBot This question has answers that may be good or bad; the system has marked it active so that they can be reviewed.
Sep 9, 2018 at 22:01 history bumped CommunityBot This question has answers that may be good or bad; the system has marked it active so that they can be reviewed.
Aug 10, 2018 at 15:02 history bumped CommunityBot This question has answers that may be good or bad; the system has marked it active so that they can be reviewed.
Jul 9, 2018 at 15:46 history bumped CommunityBot This question has answers that may be good or bad; the system has marked it active so that they can be reviewed.
Jun 1, 2018 at 12:30 comment added Martian My understanding for your case is to set bias as - average(input_values). When doing the learning, calculate the average first and set that as bias of the input layer.
Jul 3, 2017 at 14:19 comment added seanv507 So your single neuron network can never recreate the linear function y= x if you use a sigmoid. given this is just a test you should just create targets y=sigmoid (a x + b.bias) where you fix a and b and check you can recover the weights a and b by gradient descent. if you wanted to recreate the identify function, either you need an extra linear output neuron or you change to use linear neuron.
Jul 3, 2017 at 12:17 comment added ljeabmreosn @seanv507 okay, so in the future when my network contains more layers, should I add a bias term to each layer? Also, does the value of the bias term matter in this case (and its initial weight)?
Jul 3, 2017 at 8:44 comment added Thomas Wagenaar you have to backpropagate the bias value as well
Jul 3, 2017 at 7:06 comment added seanv507 you need a weight for the bias term too ( which gets adjusted by backprop too)
Jul 3, 2017 at 4:11 review First posts
Jul 3, 2017 at 5:32
Jul 3, 2017 at 4:07 history asked ljeabmreosn CC BY-SA 3.0