Latent space

So there is a very interesting technique in AI. Through the creation of adversarial neural networks we generate a mapping onto the seed of one of the networks. This mapping can be traversed to give 'in between' composites of the trained data. Such as a picture in between that of a cat and a dog. One of the most interesting features of this mapping is that arithmetic can be performed on the seeds. So if you take the man wearing glasses vector, subtract the man vector, and add the woman vector you end up with women wearing glasses.

https://www.youtube.com/watch?v=MUVbqQ3STFA