--- title: Criteria keywords: fastai sidebar: home_sidebar summary: "Which parameter is important in a neural network ?" description: "Which parameter is important in a neural network ?" nb_path: "nbs/0b_criteria.ipynb" ---
The criteria implemented come from this paper.
demo_model(random)
demo_model(large_final)
demo_model(squared_final)
demo_model(small_final)
demo_model(large_init)
demo_model(small_init)
demo_model(large_init_large_final, 80)
demo_model(small_init_small_final)
demo_model(magnitude_increase)
demo_model(movement)
The following criteria use an updating value of the weights, i.e. the value from the previous iteration of training, instead of the initialization value to better capture the training dynamics.
demo_model(updating_magnitude_increase)
demo_model(updating_movement)