GAN Support

In the past generative-adversarial NNs have shown remarkable performance increases in many fields of science. A nice summary is given in this blog article and the respective references in it.

In the framework of the ttZ analysis we can potentially profit from a GAN by generating additional background/signal events for samples for which we have low statistics instead of requesting additional MC events.

With a versatile GAN coding structure the performance of classification and regression (see #11 (closed)) NNs could potentially be improved.