Hands-On Artificial Intelligence for Beginners

Utilizing weights and biases in our MNIST example

Let's get back to our MNIST example. Let's set up our weights and biases based on the network parameters that we defined before. We'll denote our weights for connections between our input layer and first hidden layer at w1, those between our first hidden layer and second hidden layer as w2, and those between our second hidden layer and output layer as w_out. Notice how the weights track the incoming size and the outgoing size of the data at a given layer:

weights = { 'w1': tf.Variable(tf.random_normal([input_layer_size, hidden_layer_one])), 'w2': tf.Variable(tf.random_normal([hidden_layer_one, hidden_layeR_two])),  'w_out': tf.Variable(tf.random_normal([hidden_layer_two, number_classes])) ...