In dolphin-dnn, a fully connected layer can have its random generator and its random seed value for initializing its layer parameters. However, it is not needed for all fully connected layers to have their own random generators and their seed. Instead, a neural network can have a random generator and this generator can be used by all layers (initializers) including fully connected layers and other layers that will be introduced later such as convolutional layers.
In
dolphin-dnn, a fully connected layer can have its random generator and its random seed value for initializing its layer parameters. However, it is not needed for all fully connected layers to have their own random generators and their seed. Instead, a neural network can have a random generator and this generator can be used by all layers (initializers) including fully connected layers and other layers that will be introduced later such as convolutional layers.