Skip to content
Sensitivity Analysis of Deep Neural Networks (AAAI-19 paper)
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
aaaipapercodes
README.md
Slidesfor330_Shu.pdf

README.md

SA_DNN

This is the repository for the following AAAI-19 paper:

Shu, H., and Zhu, H. (In press) Sensitivity Analysis of Deep Neural Networks. The Thirty-Third AAAI Conference on Artificial Intelligence (AAAI-19). arXiv:1901.07152

Also see the talk slides.

Use CIFAR10_sample.py and MNIST_sample.py to obtain the CIFAR10 and MNIST datasets.

ResNet50.py and DenseNet121.py are the two networks, which are called to be trained by CIFAR10_ResNet50.py, CIFAR10_DenseNet121.py, MNIST_ResNet50.py or MNIST_DenseNet121.py.

Then for the two benchmark datasets, take CIFAR10 and DenseNet121 for example. Run CIFAR10_DenseNet121_IF_setupX.py for Setup X in the paper, where X=1,2,3,4. To summarize the results, first use the R code CIFAR10_DenseNet121_IF_setupX_result.R, and then use the python code CIFAR10_DenseNet121_IF_setupX_result.py for the plots.

You can’t perform that action at this time.