Certified Robustness for Top-k Predictions against Adversarial Perturbations via Randomized Smoothing

Authors: Jinyuan Jia, Xiaoyu Cao, Binghui Wang, Neil Zhenqiang Gong

ICLR 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We also empirically evaluate our method on CIFAR10 and Image Net. For example, our method can obtain an Image Net classifier with a certified top-5 accuracy of 62.8% when the ℓ2-norms of the adversarial perturbations are less than 0.5 (=127/255).Our contributions are summarized as follows:Theory. We derive the first certified radius for top-k predictions. Moreover, we prove our certified radius is tight for randomized smoothing with Gaussian noise. Algorithm. We develop algorithms to estimate our certified radius in practice. Evaluation. We empirically evaluate our method on CIFAR10 and Image Net.
Researcher Affiliation Academia Jinyuan Jia, Xiaoyu Cao, Binghui Wang, Neil Zhenqiang Gong Duke University {jinyuan.jia,xiaoyu.cao,binghui.wang,neil.gong}@duke.edu
Pseudocode Yes Algorithm 1: PREDICT; Algorithm 2: CERTIFY
Open Source Code Yes Our code is publicly available at: https://github.com/jjy1994/Certify_Topk.
Open Datasets Yes We conduct experiments on the standard CIFAR10 (Krizhevsky & Hinton, 2009) and Image Net (Deng et al., 2009) datasets to evaluate our method.
Dataset Splits No The paper mentions training and testing but does not explicitly describe a separate validation split or how it was used.
Hardware Specification No The paper does not specify any particular hardware (e.g., GPU, CPU models, or memory) used for running the experiments.
Software Dependencies No The paper does not explicitly list any software dependencies with version numbers.
Experiment Setup Yes Parameter setting: We study the impact of k, the confidence level 1 α, the noise level σ, the number of samples n, and the confidence interval estimation methods on the certified radius. Unless otherwise mentioned, we use the following default parameters: k = 3, α = 0.001, σ = 0.5, n = 100, 000, and µ = 10 5. Moreover, we use Simu EM to estimate bounds of label probabilities. When studying the impact of one parameter on the certified radius, we fix the other parameters to their default values.