Score: [0.030765441474336193, 0.9932]
Short Description of following terms:
-
Convolution : Convolution in DNN is the scanning of given input image by a feature extractor to get a reduced feature map of the image which forms a new layer in the network.
-
Filters/Kernels : Filter is a matrix of some dimension(usually 3 X 3) that extract a particular feature out of the image. It is randomly initialised by the network and is updated by back propagation if required.
-
Epochs : Epoc is a network training parameter which is described as when network is trained by the entire training data once, we call it 1 epoch.
-
1x1 Convolution : Used to reduce the number of depth channels.
-
3x3 Convolution : 3x3 Convolution matrix is most widely used convolution filter. Current hardware support is very good.
-
Feature Maps : Feature Maps are Map/collection of Features. The output of trained kernel resuls to a feature map
-
Activation Function : Activation Function is a mathematical function applied on each convolution in the network that somehow(not clear to me) impacts the output . RELU is an example of activation function.
-
Receptive Field : The receptive field is the part of the image that is visible to one filter at a time. It can be of two types: Global Receptive Fileds and Local receptive fields