is it possible to use different activations in MLP

Hi

I want to know is it possible to use different activation functions in different layers in MLP and if not why ??

also is it possible to use different activation functions in the same layer in MLP and if not why ??

last what does a gradient based algorithm means simply ?????

thanks