On Certifying Non-Uniform Bounds against Adversarial Attacks

Authors: Chen Liu, Ryota Tomioka, Volkan Cevher

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide experimental evidence in Section 4... In this Section, we compare our certified non-uniform bounds with uniform bounds. We also use our algorithm as a tool to explore the decision boundaries of different models. All the experiments here are implemented in the framework of Py Torch and can be finished within several hours on a single NVIDIA Tesla GPU machine.
Researcher Affiliation Collaboration 1EPFL, Lausanne, Switzerland 2Microsoft Research, Cambridge, UK.
Pseudocode Yes Algorithm 1 Bound Estimation
Open Source Code No The paper does not provide a direct link to the source code for the methodology described, nor does it explicitly state that the code will be made publicly available.
Open Datasets Yes real datasets, including MNIST, Fashion-MNIST and SVHN (Netzer et al., 2011).
Dataset Splits Yes 90% of the data points are in the training set and the rest are reserved for testing.
Hardware Specification Yes All the experiments here are implemented in the framework of Py Torch and can be finished within several hours on a single NVIDIA Tesla GPU machine.
Software Dependencies No The paper mentions 'Py Torch' as the framework for implementation but does not specify its version or any other software dependencies with version numbers.
Experiment Setup Yes We set the perturbation budget of PGD to be 0.1 and search for adversarial examples for 20 iterations.