During my thesis, I develop a new method, that is able to filter models used by the gradient based machine learning algorithm. It makes the model more effective. Effective model means, it only consists of general functions, those are close to the ideal model. It is useful for example when the model has more than millions parameters and we can’t develop much because of the capacity limits.
It is a generation based algorithm that observes the learning process and it makes evaluations about the parameters that belongs to the model. This is a possible solution to automatize the regularisation process of the model.
It is possible to use it in research, with the purpose of revealing new patterns in an unknown dataset. By filtering the model it opens up a new chance to improve searches to find more pattern in the dataset.
This method is an initiation of fully automatizing machine learning process. Because by automatizing the model generating and filtering process could open up the world for computers to help us understand much more coherences in the world than today.