Interpreting Adversarially Trained Convolutional Neural Networks

Authors: Tianyuan Zhang, Zhanxing Zhu

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We design systematic approaches to interpret AT-CNNs in both qualitative and quantitative ways and compare them with normally trained models.
Researcher Affiliation Academia 1School of EECS, Peking University, China 2School of Mathematical Sciences, Peking University, China 3Center for Data Science, Peking University 4Beijing Institute of Big Data Research.
Pseudocode No No pseudocode or algorithm blocks are present in the paper.
Open Source Code No The paper does not provide any statement or link indicating that the source code for the methodology is openly available.
Open Datasets Yes Three image datasets are considered, including Tiny Image Net1, Caltech-256 (Griffin et al., 2007) and CIFAR-10. 1https://tiny-imagenet.herokuapp.com/
Dataset Splits Yes Tiny Image Net has 200 classes of objects. Each class has 500 training images, 50 validation images, and 50 test images.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments.
Software Dependencies No The paper does not specify software dependencies with version numbers (e.g., specific library versions or programming language versions).
Experiment Setup Yes When training on CIFAR-10, we use the Res Net-18 model (He et al., 2016a;b); for data augmentation, we perform zero paddings with width as 4, horizontal flip and random crop. For both Tiny Image Net and Caltech-256, we use Res Net-18 model as the network architecture. We use PGD based adversarial training with bounded l and l2 norm constraints. We also investigate FGSM (Goodfellow et al., 2014) based adversarial training.