Reverse Engineering $\ell_p$ attacks: A block-sparse optimization approach with recovery guarantees

Authors: Darshan Thaker, Paris Giampouras, Rene Vidal

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on digit and face classification demonstrate the effectiveness of the proposed approach.
Researcher Affiliation Academia 1Mathematical Institute for Data Science, Johns Hopkins University, Baltimore, MD USA. Correspondence to: Darshan Thaker <dbthaker@jhu.edu>, Paris Giampouras <parisg@jhu.edu>.
Pseudocode Yes In Algorithm 1 in the Appendix, we provide the details of the active set homotopy algorithm.
Open Source Code No The paper does not provide a direct link to the source code for the described methodology or state that it is being released.
Open Datasets Yes In this section, we present experiments on the Extended Yale B Face dataset and the MNIST dataset.
Dataset Splits No The paper mentions training networks on MNIST and Yale B datasets but does not explicitly provide the train/validation/test splits, only training parameters like epochs, learning rate, and batch size.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper mentions using 'cvxpy package' with 'SCS solver' and 'Advertorch library' but does not specify their version numbers.
Experiment Setup Yes The network on MNIST is trained using SGD for 50 epochs with learning rate 0.1, momentum 0.5, and batch size 128. The ℓ PGD adversary (ϵ = 0.3) used a step size α = 0.01 and was run for 100 iterations.