In NN, the way we have different nonlinear activations, can we have different linear activations? I am just curious to understand if we can have different linear activations other than $WX+b$? I understand the necessity of weights and biases, but is this the only way out neural net's propagation works? 
Ps. I am a learner in NN, not an expert. Hence this simplistic question.
 A: Any transformation of a linear function is either a linear function with different $W$ and $b$, or it stops being linear.
That is more or less the entire reason why you need nonlinear activations in the first place - if you want to build a multi-layered model and use linear activations, e.g. $A_1(x) = 2x+1$ and $A_2(x) = 3x$ in a two-layer model, then you might as well replace them with a single layer doing $A_{1+2}(x) = 6x+3 = 3*(2x+1)$.
A: There is no such thing as other linear activations but other nonlinear activations, such as sigmoid, tanh, ReLU, leaky ReLU etc, which are typically used to equip the network with different functionalities. All linear activation will provide no additional capability to the network because all layers can be reduced into one (as also pointed out in the other answer). 
P.S. There isn't a total consensus on use of linear/affine functions. $f(x)=ax+b$ is denoted as an affine function (where $f(x)=cx$ is denoted as linear) but in many places it's also noted as linear function.
