Ti-Lin

Codes for reproducing the experimental results in "Certifying Robustness of Convolutional Neural Networks with Tightest Linear Approximation".

This repository contains our tools Ti-Lin and Ti-Lin-R. Concretely, for general CNNs/FNNs we compare Ti-Lin's linear bounds of Sigmoid/Tanh/Arctan function with VeriNet and DeepCert, and compared Ti-Lin's linear bounds of Maxpool function with CNN-Cert. For non-negative CNNs, we compared Ti-Lin-R with VeriNet. The experiment results show that Ti-Lin gains tighter non-trivial certified robustness bounds with almost the same computational complexity.

Setup

1. The code is tested with python3 and TensorFlow v1.10 or v1.12 (TensorFlow v1.8 is not compatible). The following packages are required.

   conda create --name tilin python=3.6
   conda activate tilin
   conda install pillow numpy scipy pandas h5py tensorflow=1.12.0 numba posix_ipc matplotlib
   

2. Clone this repository

   git clone https://github.com/xiaobailearn/Ti-Lin.git
   cd Ti-Lin
   

3. Download the models in this paper and unzip the zip file.

   unzip models.zip
   

4. Reproduce the experiments result in our paper.

• In DeepCert, VeriNet, and Ti-Lin, you can get certified robustness bounds and runtime on pure CNNs with Sigmoid/Tanh/Arctan functions in two steps. Firstly, you need to set table in the main function of pymain_new_network.py to 1. Secondly, you need to run pymain_new_network.py. The result are saved into log_pymain_new_network_{timestamp}.txt. For example,

  cd Ti-Lin
  python pymain_new_network.py 
  

• In DeepCert, VeriNet, and Ti-Lin, you can get certified robustness bounds and runtime on FNNs with Sigmoid/Tanh/Arctan functions by setting table in the main function of pymain_fnn.py to 2 and running pymain_fnn.py. For example,

  cd Ti-Lin
  python pymain_fnn.py 
  

• In Cnncert and Ti-Lin, you can get certified robustness bounds and runtime on CNNs with max pooling with ReLU by setting table in the main function of pymain_new_network.py to 3 and running pymain_fnn.py. For example,

  cd Cnncert
  python pymain_new_network.py 
  

• In VeriNet and Ti-Lin-R, you can get certified robustness bounds and runtime on non-negative pure CNNs with Sigmoid/Tanh/Arctan functions by setting table in the main function of pymain_new_network_postive.py to 4 and running pymain_new_network_postive.py.

  cd VeriNet
  python pymain_new_network_postive.py 
  

  Results of certified robusntess bounds and runtime are saved into log_pymain_new_network_{timestamp}.txt

References

Boopathy, A.; Weng, T.-W.; Chen, P.-Y.; Liu, S.; and Daniel,L. 2019. Cnn-cert: An efficient framework for certifying robustness of convolutional neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, 3240–3247.