Improving l1-Certified Robustness via Randomized Smoothing by Leveraging Box Constraints
Authors: Vaclav Voracek, Matthias Hein
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We performed an extensive evaluation on CIFAR10 (Krizhevsky et al., 2009) and Image Net-1k (Russakovsky et al., 2015) |
| Researcher Affiliation | Academia | V aclav Vor aˇcek 1 Matthias Hein 1 1 T ubingen AI Center, University of T ubingen. Correspondence to: V aclav Vor aˇcek <vaclav.voracek@uni-tuebingen.de>, Matthias Hein <matthias.hein@uni-tuebingen.de>. |
| Pseudocode | Yes | Algorithm 1 Randomized Smoothing Certification |
| Open Source Code | Yes | Code is released at https://github.com/vvoracek/L1-smoothing. |
| Open Datasets | Yes | We performed an extensive evaluation on CIFAR10 (Krizhevsky et al., 2009) and Image Net-1k (Russakovsky et al., 2015) |
| Dataset Splits | Yes | For CIFAR-10 dataset we certify 2 000 images from the test set, while for Image Net we certify the same subset of 500 images as Cohen et al. (2019) and Levine & Feizi (2021). |
| Hardware Specification | No | No specific hardware details (like GPU models, CPU types, or memory) used for running the experiments are mentioned in the paper. |
| Software Dependencies | No | The paper mentions 'Python function proportion confint from package statsmodels.stats.proportion' but does not provide specific version numbers for this or any other software dependency. |
| Experiment Setup | Yes | For Image Net we used a Res Net-50 trained for 30 epochs; and for CIFAR-10 we used a Wide Res Net-40-2 trained for 120 epochs. The optimizer is SGD with learning rate 0.1, Nesterov momentum 0.9 and weight decay 0.0001 with cosine annealing learning rate schedule and batch size is 64 for both models. |