Consider a neural network in matrix$-$vector notation.

Assume that the preactivation vector for layer $l-1$ is given by $s^{[l-1]}=(-2,-1,3{)}^T.$ As activation function $f$ take Relu.

The weight matrix reads $W^{[l]}=([1,2,3],[2,3,4],[3,4,5]).$

Compute the bottom right entry of the single$-$layer Jacobian $J^{[l]},$ i$.$e$.J_{3,3}^{[l]}.$

The answer is an integer number.

Antwort: