4 ms·
> the training process can be seen as a giant matrix multiplication in one step. It can't be seen that way. Deep learning becomes deep when you alter linear m
by nabla9 8y ago
> the training process can be seen as a giant matrix multiplication in one step.
It can't be seen that way.
Deep learning becomes deep when you alter linear matrix multiplications with non-continuous functions. After that you can't reduce the problem into one step. You also lose associativity so you can't even do matrix chain multiplication optimization.