22  Mode Collapse

Mode collapsing happens when the generator produce only a subset of the patters or modes found in the original distribution.
In our case all outputs are persons which have as event DEATH.

To understand why the model is mode collapsing I should describe in general how a GAN works

The idea is to use Generator to create a vector on the latent space generated by the Encoder and use the Decoder to convert those values which will then be compared, by the discriminator, to an original sample.

The issue is, many times, the Discriminator is much more powerful than the Generator: the Generator won’t have time to learn all the features before the Discriminator became too good and prevents the Generator to further improve its performance.

Obviously I didn’t want the GAN to mode collapse so I tried find solutions