asked 2017-12-17 10:44:25 -0500

sjhalayka gravatar image

updated 2017-12-17 11:17:44 -0500

What’s the main difference between the ANN_MLP and the DNN? Do both provide artificial neural networks? Is the only difference between a regular ANN and a deep network is that the deep network has multiple hidden layers?

edit retag flag offensive close merge delete


the ann has only one type of layers -- fully connected ones (think of it as a matrix-multiplication)

while dnns can contain a multitude of other concepts, convolutional layers, pooling, even layers, that are small ann's themselves (like inception)

berak gravatar imageberak ( 2017-12-17 12:59:16 -0500 )edit

Right on. Thanks @berak

sjhalayka gravatar imagesjhalayka ( 2017-12-17 13:23:27 -0500 )edit

Is there a standard text for DNN applications?

sjhalayka gravatar imagesjhalayka ( 2017-12-17 16:49:12 -0500 )edit

I have read the Wikipedia article on deep neural networks, and it says that multiple hidden layer ANNs are considered to be deep neural networks. I tried to make a multiple hidden layer ANN using ANN_MLP, and it seems to work (well, no assertions fail nor are exceptions thrown).

const int num_input_neurons = dove.cols;
const int num_output_neurons = 2;
const int num_hidden_neurons = static_cast<int>(sqrtf(image_width*image_height*num_output_neurons));
Mat layersSize = Mat(4, 1, CV_16UC1);
layersSize.row(0) = Scalar(num_input_neurons);
layersSize.row(1) = Scalar(num_hidden_neurons);
layersSize.row(2) = Scalar(num_hidden_neurons);
layersSize.row(3) = Scalar(num_output_neurons);
sjhalayka gravatar imagesjhalayka ( 2017-12-17 17:17:23 -0500 )edit

yes, above should work, but that's probably already the limit, what you can do with it.

and i don't think, there's a standard text for deep learning .. here are some links, i found useful:

berak gravatar imageberak ( 2017-12-18 01:29:00 -0500 )edit

Thanks @berak !

sjhalayka gravatar imagesjhalayka ( 2017-12-18 10:24:21 -0500 )edit