Certifying Confidence via Randomized Smoothing

Authors: Aounon Kumar, Alexander Levine, Soheil Feizi, Tom Goldstein

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experimental results on CIFAR-10 and Image Net datasets show that using information about the distribution of the confidence scores allows us to achieve a significantly better certified radius than ignoring it.
Researcher Affiliation Academia Aounon Kumar University of Maryland aounon@umd.edu Alexander Levine University of Maryland alevine0@cs.umd.edu Soheil Feizi University of Maryland sfeizi@cs.umd.edu Tom Goldstein University of Maryland tomg@cs.umd.edu
Pseudocode No The paper describes its methods in prose and does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Code for the experiments is available at https://github.com/aounon/cdf-smoothing.
Open Datasets Yes Our experimental results on CIFAR-10 and Image Net datasets show that using information about the distribution of the confidence scores allows us to achieve a significantly better certified radius than ignoring it.
Dataset Splits No The paper mentions using ResNet models trained by Cohen et al. in [7] on CIFAR-10 and Image Net datasets, but it does not explicitly state the training, validation, or test dataset splits used for its own experiments.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., CPU, GPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific version numbers for ancillary software dependencies (e.g., libraries, frameworks) used in the experiments.
Experiment Setup Yes We use the same number of samples m = 100, 000 and value of α = 0.001 as in [7]. We set s1, s2. . . . , sn in theorem 2 such that the number of confidence score values falling in each of the intervals (a, s1), (s1, s2), . . . , (sn, b) is the same. We use the same σ for certifying confidences as well.