Which Python Architecture to use for accessing existing algorithm???(keras, Tensorflow_privacy, PyTorch etc)

Which DL Model to use?(like CNN, CDBN, DNN, LSTM etc)

Is there any Privacy Preserving Pretrained Model Available??

Is it good to use Pre-trained model??(like VGG16, ResNet, ResNeXT,ResNet50)

Which Privacy Mechanism To use?( Differential Privacy, Homomorphic Encryption, K-anonymity etc)

Where to Provide Privacy? at Data Level or Model level ??(Data Pre-processing, Model weight Privacy, Privacy between layers, Privacy at activation function)

How to work with Federated Learning?? (is it good approach, computationally feasible)

How to Provide Privacy in Federated Learning ??(which privacy method to use)

How to distribute noise in differential Privacy??(Apart from Laplacian/Gausian distribution, any other effective noise distribution??)

How much amount of noise we need to add?? (Not to loose data utility...)