How are the weights and biases are updated after training the MLP?

Question

I\'m confused about something in the code below of MLP tutorial using python TensorFlow. The function def mlp(x) is called twice in preds = tf.nn.softmax(