I already post the question on Stackoverflow : https://stackoverflow.com/questions/53785922/is-the-use-of-a-non-monotonic-activation-function-is-a-correct-and-viable-solut?noredirect=1#comment94423971_53785922
I found several papers about how to build a single layer perceptron able to solve the XOR problem. The papers describe a solution where the heaviside step function is replaced by a non-monotonic activation function. Here is the papers :
I also found this topic on Stackoverflow : https://stackoverflow.com/questions/30412427/solving-xor-with-single-layer-perceptron
Can we really solve the XOR problem with a simple perceptron ? Because the use of a non-monotonic activation function is not really common on this topic. Papers about this idea are scarce. Generally, the main solution is to build a multilayer perceptron.