IBM / CROWN-Robustness-Certification

CROWN: A Neural Network Verification Framework for Networks with General Activation Functions
Apache License 2.0
38 stars 11 forks source link

version of pytorch #1

Open miaoxiaodaiblack opened 3 years ago

miaoxiaodaiblack commented 3 years ago

Hi! Your work is so interesting! Thank you! But is there the code written by pytorch?

huanzhang12 commented 3 years ago

@miaoxiaodaiblack Thank you so much for your interests in our work!

In our CROWN-IBP repository we provide a PyTorch implementation of CROWN on simple convolutional neural networks . See instructions here.

Moreover, we developed the auto_LiRPA library which implements CROWN in PyTorch for general neural network architectures. We support CROWN based certified training and verification on fairly complicated networks like ResNeXt, DenseNet, LSTM and Transformer. The library is user friendly and easy to use: you only need to define a forward() function in PyTorch and you can obtain bounds automatically without any manual derivations. We provide many examples for our library and you can also checkout our Colab Demo for a quick start.

Both libraries are very efficient and run on (multi-)GPUs with PyTorch, and can handle relatively large networks. Feel free to let me know if you need any additional help!

Thanks, Huan Zhang

miaoxiaodaiblack commented 3 years ago

Thank you for your reply!And I have another confusion about the auto-LIPRA and looking forward for your reoply!Sent from my Huawei phone-------- Original message --------From: Huan Zhang @.>Date: Sat, Mar 6, 2021, 6:20 PMTo: IBM/CROWN-Robustness-Certification @.>Cc: miaoxiaodaiblack @.>, Mention @.>Subject: Re: [IBM/CROWN-Robustness-Certification] version of pytorch (#1) @miaoxiaodaiblack Thank you so much for your interests in our work! In our CROWN-IBP repository we provide a PyTorch implementation of CROWN on simple convolutional neural networks . See instructions here. Moreover, we developed the auto_LiRPA library which implements CROWN in PyTorch for general neural network architectures. We support CROWN based certified training and verification on fairly complicated networks like ResNeXt, DenseNet, LSTM and Transformer. The library is user friendly and easy to use: you only need to define a forward() function in PyTorch and you can obtain bounds automatically without any manual derivations. We provide many examples for our library and you can also checkout our Colab Demo for a quick start. Both libraries are very efficient and run on (multi-)GPUs with PyTorch, and can handle relatively large networks. Feel free to let me know if you need any additional help! Thanks, Huan Zhang

—You are receiving this because you were mentioned.Reply to this email directly, view it on GitHub, or unsubscribe.

miaoxiaodaiblack commented 3 years ago

Thank you for your reply!