search

laura-rieger / deep-explanation-penalization

Code for using CDEP from the paper "Interpretations are useful: penalizing explanations to align neural networks with prior knowledge" https://arxiv.org/abs/1909.13584
MIT License
127 stars 14 forks source link

What's the performance when CNN is also trained? #6

Closed Juna2 closed 4 years ago

Juna2 commented 4 years ago

Hi, Laura. I found that with the vgg16 you trained with the ISIC dataset, actual training only happens on the FC layer at the end. Have you tried training all the layers as well? Thanks in advance!

laura-rieger commented 4 years ago

Hi, we tried training the entire net (without CDEP or any other regularization) on the ISIC dataset. The performance was below training only the FC layers and the model overfit, presumably because the dataset is relatively small for training an entire vgg16. Therefore we focused on finetuning the last layers in our experiments.

Juna2 commented 4 years ago

Thank you for the information! Have a good day!