Why do we not need to initialize parameters when using tensorflow

In the code used from video 1 to video 3 of week 2, at no point did we initialize the weights and biases? How are we getting by without doing it? Can tensorflow do that too by itself?

You don’t need to do it explicitly, TensorFlow initializes the parameters for you. However, you can change the default behavior of TF to initialize them differently as you want…