Differences

This shows you the differences between two versions of the page.

--- data_mining:neural_network:deep_neural_nets [2017/08/20 19:25] – [Backwad prop] phreazer
+++ data_mining:neural_network:deep_neural_nets [2017/08/20 20:04] – [Matrix dimensions] phreazer
@@ Line 24: / Line 24: @@
 $A^{[0]}$ is input set.
-===== Backwad prop =====
+===== Backward prop for layer l =====
 Input: $da^{[l]}$
@@ Line 31: / Line 31: @@
-$dz^{[l]} = da^{[l]} * g'^{[l]}(z^{[l]})$
+$dZ^{[l]} = dA^{[l]} * g'^{[l]}(Z^{[l]})$ # element-wise product
-$dW^{[l]} = dz^{[l]} * a^{[l-1]}$
+$dW^{[l]} = 1/m * dZ^{[l]} * A^{[l-1]^T}$
-$db^{[l]} = dz^{[l]}$
+$db^{[l]} = 1/m * np.sum(dZ^{[l]}, axis=1, keep.dims=True)$
-$da^{[l-1]} = W^{[l]}^T * dz^{[l]}$
+$dA^{[l-1]} = W^{[l]^T} * dZ^{[l]}$
+===== Flow =====
+Forward:
+X -> ReLU -> ReLU -> Sigmoid -> $\hat{y}$ -> $L(\hat{y}, y)$
+Init backprop with derivative of $L$.
+===== Matrix dimensions =====
+$l=5$
+-3-5-4-2-1
+$Z^1 = W^1 * x + b^1 $
+$Z^1 :(3,1)$
+$x : (2,1)$
+$W^1 :(n^1,n^0) => W^1 (3,2), W^2(5,3)$
+$W^l :(n^l, n^{l-1})$
+$b^1 : (3,1)$
+$b^L : (n^l, 1)$
+analog with $dW^l$ and $db^l$
+==== Vectorized ====
+$Z^1 : n^1,m$
+$W^1 :(n^1, n^0)$
+$X : (n^0, m)$
+$b^1: (n^1,m)$