/sci/ - Science & Math

>>7720322
they're really not as mysterious as they sound.

>arrange your input data into a vector(eg a vector of pixel intensities for an image)
>multiply the vector by a weght matrix to get the input matrix for the first hidden layer
>evaluate a function(eg tanh(x)) on every element of that matrix to get the activation matrix
>multiply this by the second weight matrix to get the second hidden layers input.
>repeat for as many hidden layers as you want

congratulations, you made a feedforward neural network!( training it with back propagation is only slightly harder.)

'deep learning' isn't all that arcane either. 'deep' just means you have a lot of hidden layers