09.05.2023 Views

pdfcoffee

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

The Math Behind Deep Learning

In this case the sum will not disappear because the change of weights in the hidden

layer is directly affecting the output. Substituting yy oo = δδ oo (zz oo ) and applying the

chain rule, we have:

∂∂∂∂

= ∑(yy

∂∂ww oo − tt oo ) ∂∂δδ oo (zz oo )

iiii

oo

∂∂ww iiii

= ∑(yy oo − tt oo )δδ′ oo (zz oo ) ∂∂∂∂ oo

∂∂∂∂ iiii

oo

The indirect relation between z o

and the internal weights w ij

(Figure 13) is

mathematically expressed by the expansion:

zz oo = ∑ ww jjjj δδ jj (zz jj ) + bb oo =

jj

∑ ww jjjj δδ jj (∑ ww iiii zz ii + bb ii ) + bb oo

jj

ii

since zz jj = ∑ ww iiii zz ii + bb ii

This suggests applying the chain rule again:

ii

∂∂zz oo

∂∂ww iiii

= (chain rule)

= ∂∂zz oo

∂∂yy jj

∂∂yy jj

∂∂ww iiii

= (substituting z 0

)

= ∂∂yy jjww jjjj

∂∂yy jj

∂∂yy jj

= (deriving)

∂∂ww iiii

= ww jjjj

∂∂yy jj

∂∂ww iiii

= (substituting yy jj = δδ jj (zz jj ) )

= ww jjjj

∂∂δδ jj (zz jj )

∂∂ww iiii

= (chain rule)

= ww jjjj δδ′ jj (zz jj ) ∂∂∂∂ jj

= (substituting zz jj = ∑ yy ii ww iiii + bb ii

∂∂∂∂ iiii

ii

[ 558 ]

)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!