This is an experiment to achieve resampling of point sets in 2D using a neural network. The network is trained in an unsupervised way, with only the spectral characteristic curve. The training set is initialized with random Gaussian data, and using a simple MLP the point set can be moved towards Poisson distribution.
However, this is not ideal since there is always a bias towards learning a same representation and the output lacks variability and randomness.
