Deep Learning Models

Models:

Following models were considered for the PoC in the original experiments. However, the scope of this project is not limited to these models alone. Any new model considered should be added to the model directory in the project along with the performance evaluation report in the analysis directory.

Feed forward Deep Neural Network:

Model 1:

A 5 layerd DNN with batch normalization and reLU activation in the hidden layers:

Model 2:

A 5 layered DNN with batch normalization and biased sigmoid activation in the hidden layers:

Model 3:

A CNN with convolution2dLayers and using downsampling by the maxPooling2dLayer and leaky reLU activation of scale 0.01. The convolutional filter’s kernel size was kept at 11x11 and the number of convolutional filters was kept the same as feature size i.e 64/121 for the first convolution2dLayer, 2x(feature size i.e 64/121) for the second convolution2dLayer and 4x(feature size i.e 64/121) for the remaining two convolution2dLayer(s).