Question

Assume the artificial neural network below, with mean square error loss and true output of 3. All activation functions are ReLU. Perform one round of forward propagation and calculate the loss. Then perform a Stochastic Gradient Descent update using back propagation to get the new values of weights $W_1$ and $W_7$ with learning rate 0.01. input layer ReLU layer output layer $w_1 = 2$ $w_2 = 1$ $w_7 = -1$ $x_1 = 1$ $w_3 = 1$ $w_8 = 3$ $w_4 = -2$ $x_2 = 2$ $w_5 = 1$ $w_9 = 2$ $w_6 = 2$ You can work out the problem on paper and take photos of your calculation process.

          Assume the artificial neural network below, with mean square error loss and true output of 3.
All activation functions are ReLU.
Perform one round of forward propagation and calculate the loss.
Then perform a Stochastic Gradient Descent update using back propagation to get the new values of weights $W_1$ and $W_7$ with learning rate 0.01.
input layer
ReLU layer
output layer
$w_1 = 2$
$w_2 = 1$
$w_7 = -1$
$x_1 = 1$
$w_3 = 1$
$w_8 = 3$
$w_4 = -2$
$x_2 = 2$
$w_5 = 1$
$w_9 = 2$
$w_6 = 2$
You can work out the problem on paper and take photos of your calculation process.

Assume the artificial neural network below, with mean square error loss and true output of 3.
All activation functions are ReLU.
Perform one round of forward propagation and calculate the loss.
Then perform a Stochastic Gradient Descent update using back propagation to get the new values of weights W1 and W7 with learning rate 0.01.
input layer
ReLU layer
output layer
w1 = 2
w2 = 1
w7 = -1
x1 = 1
w3 = 1
w8 = 3
w4 = -2
x2 = 2
w5 = 1
w9 = 2
w6 = 2
You can work out the problem on paper and take photos of your calculation process.

Added by Matthew U.

Question

Please give Ace some feedback