Center Smoothing: Certified Robustness for Networks with Structured Outputs

Authors: Aounon Kumar, Tom Goldstein

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our method to create certifiably robust models with disparate output spaces from sets to images and show that it yields meaningful certificates without significantly degrading the performance of the base model.
Researcher Affiliation Academia Aounon Kumar University of Maryland aounon@umd.edu Tom Goldstein University of Maryland tomg@cs.umd.edu
Pseudocode Yes Algorithm 1 Smooth Algorithm 2 Certify
Open Source Code Yes Code is included in the supplemental.
Open Datasets Yes We use a pre-trained face detection model for this experiment... on the Celeb A face dataset [45]... a generative adversarial network Big GAN pre-trained on Image Net images [5]... MNIST [16] and CIFAR-10 [32]. We are using datasets that are available in the public domain with custom license terms that allow non-commercial use, like MNIST, CIFAR-10 and Celeb A.
Dataset Splits No The paper mentions using 'n' and 'm' samples for smoothing and certification respectively, and training on noisy data with 'σtrain', but it does not specify explicit train/validation/test dataset splits with percentages or counts for reproducibility.
Hardware Specification Yes We ran all our experiments on a single NVIDIA Ge Force RTX 2080 Ti GPU in an internal cluster.
Software Dependencies No The paper does not specify any software names with version numbers (e.g., programming languages, libraries, or frameworks like Python, PyTorch, TensorFlow, etc.) used for the experiments.
Experiment Setup Yes We train our base models (except for the pre-trained ones) on noisy data with different noise levels σtrain = 0.1, 0.2, . . . , 0.5... We use n = 104 samples to estimate the smoothed function and m = 106 samples to generate certificates, unless stated otherwise. We set = 0.05, α1 = 0.005 and α2 = 0.005... We set n = 5000 and m = 10000, and use default values for other parameters discussed above.