The Back-propagation algorithm
The back-prop algorithm for supervised learning
from exemplars is:
Repeat:
-
Send in inputs Ii for this exemplar.
-
Calculate outputs yk
-
Get given correct outputs Ok
-
Measure E.
- wjk weights:
-
Calculate all the
=
yk ( 1 - yk ) ( yk - Ok )
-
Calculate all the
=
yj
-
For all j,k:
- wij weights:
-
Calculate all the
=
yj ( 1 - yj )
Σ k
(
wjk
)
-
Calculate all the
=
Ii
-
For all i,j:
- Repeat (next exemplar).
See sample code.