use adjusted GAN architecture to aid (EEG) data blackening

[neurodream]

rationale: data sharing whilst making sure that subjects are not identifiable

parts of system:

• data identificator tries to identify subjects from data
• data transformer tries to adjust data so that identificator runs at chance performance
• data usability checker checks whether the transformed data still contain established statistical properties (ERPs, etc.)