Learning curly presentations from the dear data is a basic challenge for a machine learning. Resolution can unlock other problems, such as usual, interpretation, or equality. Although it is a wonderful challenge to solve the idea, the demolition is often benefited from achieving previous similarities. In addition, the latest jobs have shown that similar methods can be developed with geometrical, eg. By reading the representations that keep the data features of data, such as distance or angles are between points. However, the previous similarity while maintaining gomeriar features is a challenge, as a fully maintained map of these features while adapting the data distribution through the order. Dealing with these challenges, introducing a way of an illegal learning based on excellent transportation. We create a problem using Grame-Mon maps that move one distribution to other little crashes of previously defined geometric features, save as soon as possible. To join such maps, we raise the Grame-Monge-Monge-Gap (GMG), the Stigatizer calling no matter removes the distribution distribution of the geometric distortion. We reflect the effectiveness of our approach to all four familiar benches, reduce other means affecting geometric consideration.
* Equal donation
** Equal advice
40 CREST-Ensae
‡ Helmholtz Munich
§Tuu Munich
¶Cml
Eur4 40 Tubingen Ai Center
