Week4 Building Blocks of Deep Neural Networks

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

We store cache (z) for every neuron. Is no of cache depends on no of units??

2

Hi PrasadChaskar,

In general, when caches are used, they are used to store values needed for backpropagation. For a standard neural network, this includes all values of Z that depend on the activations of the neurons. So per neuron, a value of z needs to be stored to be able to calculate the gradients in backpropagation.

1 Like
3

Of course the key point implicit in what Reinoud points out here is that the Z and A values stored in the caches are vectors or matrices (as appropriate), so that you don’t need more caches. All the per neuron values are contained in the array representations of the values.