Vous êtes sur la page 1sur 1

Widrow-Hoff Training Method http://www.cse.msu.edu/~cse449/spr98/lectures/week5/tsld022.

htm

Widrow-Hoff Training Method(Incremental)


1. Set the weights to small random values (e.g. between -.1 and .1)

2. Set the learning rate a << 1.

3. Repeat // until training error is low enough

Set total squared error = 0;

For each training example e

begin

for j=0 to N do

w[j] = w[j] + a*(d[e] - o[e])*x[j,e]; // x[0,e] is always 1

error = error + square(d[e] - o[e]);

end

until error < desired_value.

4. Store weight vector w in a file.

Previous slide Next slide Back to first slide View graphic version

1 sur 1 05/03/2011 21:13

Vous aimerez peut-être aussi