Clarifications of some aspects related to neural network incremental training

From the Mathworks documentation I understand that there are two training functions available for incremental learning:

1. trainc for cyclic training ( https://uk.mathworks.com/help/deeplearning/ref/trainc.html )
2. trainr for random training ( https://uk.mathworks.com/help/deeplearning/ref/trainr.html )

From the information provided at https://uk.mathworks.com/help/deeplearning/ug/neural-network-training-concepts.html I also understand that incremental training is triggered automatically when input patterns are provided as cell arrays rather than matrices.

My questions related to these topics are as follows

1. Is my understanding of the incremental learning triggering correct? Does one need to provide the network input as cell arrays to use incremental learning, or is it also possible to provide them as a matrix and set trainFcn as one of trainc or trainc instead?
2. Does trainr operate randomization with or without replacement?
3. For incremental training, can a different learning rate be tagged to each individual training pattern?
4. Which kind of networks work with the incremental training trainc and trainr functions?

Many thanks in advance for your answer.