Backpropagation alogorithm
what is δ³?
(theta³)'δ⁴.*g'(z³) where g'(z³) = a³.* (1- a³)
how do you compute the gradient?
1) do forward propagation on inputs. This calculates all activation points values. 2) followed by back propgation, calculating delta at output layer, followed by all other layers
Backpropagation alogorithm?
calc delta at ouput layer and work backwards
what is ∆^l∨ij?
contains series of delta. used to compute partial derivative
what does δ^l∨j compute?
error of activation unit in layer l and unit j. ie a^l∨j - y∨j
notation? θ∨ij∧l ∈ R
i activation unit j layer l layer
do we calculate delta¹
no because we don't associate error term with input layer
what the vectorized formula for δ^l∨j in final layer?
δ⁴=a⁴-y