Is there a way to remove the layers of a pre-trained Alexnet trained on the places databse, and test on the remaining network?
What I would like to do is study the impact of removing one layer at a time in order to see how that impacts model performance. This will allow me to see which layers are the most important to specific categories of scenes in the places database.