How to Pinch a Neural Network

There are more and more sophisticated techniques, pruning techniques for neural networks. For example, the first convolutional layers of a competer vision nural network have much less parometers with respect to the last layers. So you know, one should take into account this diversity and this big difference in number of parometers for for pruning. It's not formally guaranteed that in finding these tickets is consistent across different metodologies, different data sets, different apologies, et cetera.