C3W2 Assignment - Question about input batches

Hi everyone,

In C3W2 assignment, Exercise 02, the data generator returned two copies of the same input batches, the first used for inputs and the second used for targets. My question is, wouldn’t this encourage the neural network to generate things like y=x (one that always gives a same output as input)? Given that the neural network was not doing that, why is this?

Thanks in advance.

Hi @mechbear14

This is a good question and unfortunately it is somewhat never clearly explained in the Course.

The neural network (model) never sees the second part, it only sees the x (inputs).

The second part is used in training / calculating the loss. In other words, neural network takes the inputs → generate predictions (outputs) and these outputs are compared with “targets” (the second part). Then the weights of the model are updated accordingly.

Cheers