How to design a locally connected layer for use in a convolution neural network??
1 Ansicht (letzte 30 Tage)
Ältere Kommentare anzeigen
I am trying to replicate the architecture found in this paper (https://www.nature.com/articles/srep36571), but the deepNetworkDesigner app does not have a locally connected layer. The best alternative I found is to use the 2d convolution but with 1x1 filter size to approximate the 1d convolution behavior (https://stackoverflow.com/questions/50388014/1d-convolution-for-cnn). According to (https://keras.io/api/layers/locally_connected_layers/locall_connected1d/) the locally connected layer is similar to 1d convolution, except the weights are unshared.
How can I go about actually making a proper 1d convolution layer?
0 Kommentare
Antworten (1)
Srivardhan Gadila
am 30 Okt. 2020
You can refer to Define Custom Deep Learning Layers & Deep Learning Custom Layers and implement your own custom deep learning layer.
Siehe auch
Kategorien
Mehr zu Sequence and Numeric Feature Data Workflows finden Sie in Help Center und File Exchange
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!