Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness

Authors: Vaclav Voracek

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide empirical validation of the proposed methods confirming the theory. ... We benchmark the confidence sequences on the Sequential decision making task, where we try to certify a certain radius at given confidence level with as few samples as possible... Cifar10, ℓ2, details are in Appendix C.2.1
Researcher Affiliation Academia Václav Voráˇcek Tübingen AI center, University of Tübingen vaclav.voracek@uni-tuebingen.de
Pseudocode Yes Algorithm 1 Union-Bound Confidence Sequence... Algorithm 2 Betting Confidence Sequence
Open Source Code Yes The code can be found at https://github.com/vvoracek/RS_conf_seq.
Open Datasets Yes right: Comparison of ℓ2robustness curves with the standard (dashed) or the randomized (solid) Clopper-Pearson bounds on a CIFAR-10 dataset under the standard setting. The experimental details are in Appendix C.
Dataset Splits No The paper mentions using 'CIFAR-10 dataset' and refers to 'test dataset' in its experimental details (Appendix C.1) but does not explicitly state the training/validation/test dataset splits, such as percentages or specific counts for each partition, nor does it explicitly mention the use or creation of a validation set beyond the standard test set.
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, processor types, or memory amounts used for running its experiments. It only vaguely mentions 'single GPU' in the NeurIPS checklist.
Software Dependencies No The paper does not explicitly list specific software dependencies with their version numbers, such as 'PyTorch 1.9' or 'CUDA 11.1', which are needed to replicate the experiments.
Experiment Setup Yes We had Wide Res Net-40-2 for CIFAR-10 trained for 120 epochs with SGD and learning rate 0.1, Nesterov momentum 0.9, weight decay 0.0001 and cosine annealing. batch size 64.