I am trying to create a network with dynamic connections for every different training example

Question

I have developed a new algorithm which has dynamic connections. it is based off of the following paper:

https://www.researchgate.net/publication/334226867_Dynamic_Encoder_Network_to_Process_Noise_into_Images

i have figured out how to effect the dynamism. The pseudo code is in the diagram.

Additionally i would really like to know how to do the rest of the code in terms of a general overview. i would need to have the same deconvolutional layer for all training examples, as well as different current_weight_matrices for the different training examples. so how do i go about adding the dynamic layers to the deconv neural network . Will i have to have many copies of the same network?