On: Are deep neural networks dramatically overfitted? (2019)

[yxhuvud]

Well it is not the first time the idea of dimensionality reduction have been used in the field. It is for example the idea behind latent semantic methods (LSI and LSA).