To design a Multilayer Perceptron, should I use more units per layer and less layers or more layers and less units, which is better?

Question

I feel confused about how to design a Multilayer Perceptron(MLP). I have read a number of papers about how to design a deep CNN structure, like VGG or ResNet. However, for a Multilayer perceptron, a more basic ANN structure, should I follow this trend to make the MLP deeper? Actually, I try some trials on python that stacking more layers in one network, but the result showed it make no sense.

Therefore, I just wanna ask the question above.

can anybody give me some suggestions?

score 1 · Answer 1 · edited Apr 13 '17 at 12:44

1

The optimal neural network architecture depends on your data set.

FYI:

edited Apr 13 '17 at 12:44

Community

1

answered Nov 09 '16 at 17:49

Franck Dernoncourt

42,093
30
155
271

1

One more: https://stats.stackexchange.com/questions/182734. – amoeba Apr 19 '17 at 14:27

To design a Multilayer Perceptron, should I use more units per layer and less layers or more layers and less units, which is better?

1 Answers1