About the hidden layers and what features they extract

Would the hidden layers always extract useful features like edges and face shapes in a way a human might identify these features being useful or maybe at times extract features even in the higher levels that may not make any sense to a human. If so then I think it’s really fascinating that an artificial intelligence is not a dumbed-down version of human intelligence but instead is something completely new.

1 Like

Hi @Pranil_Dhakal

In course 2 week2, there is an assignment on using neural network algorithm to identify hand-written digits. So hold your thought and you would be able to answer that question yourself.

Hi, @Pranil_Dhakal !

Neural networks usually extract smaller features in the first layers and then bigger ones on the last ones. That happens because the receptive field increase throughout the network, so fine details like edges are detected at the input layers and more coarse ones, like faces, objects, are observed near the output.