Reversible learning of artificial neural networks

Hyperparameter tuning is one of the main challenges that has to be tackled in order to create an efficient machine learning model. The optimal values of the hyperparameters differ depending on the datasets and the tasks that have to be performed on them. This thesis approaches the topic of gradient-based hiperparameter tuning, using the reversibility of the learning process conducted with the Classic Momentum algorithm. This feature enables the recreation of the learning trajectory and memory-efficient hyperparameter gradient computation. Obtaining such gradient facilitates the construction of more powerful models. The implementation of this solution is going to be presented in the Python programming language with the inclusion of the Tensorflow library.