A end to end differentiable neural network able to sample from any probability distribution, such as the exponential distribution.
The major requirement is to have a sufficiently larger number of realizations, drawn from the target distribution.
The neural sampler is controlled by only one parameter: the seed. It controls the randomness in the generation. For two identical seeds, the generated realizations should match exactly.
NB: This first implementation restricts the sampling to univariate random variables.
Normal Distribution -> Exponential Distribution (during training phase)