Prediction Poisoning: Towards Defenses Against DNN Model Stealing Attacks

Authors: Tribhuvanesh Orekondy, Bernt Schiele, Mario Fritz

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

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments, we find our approach consistently mitigate various attacks and additionally outperform baselines.
Researcher Affiliation Academia Tribhuvanesh Orekondy1, Bernt Schiele1, Mario Fritz2 1 Max Planck Institute for Informatics 2 CISPA Helmholtz Center for Information Security Saarland Informatics Campus, Germany
Pseudocode Yes We further elaborate on the solver and present a pseudocode in Appendix C.
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of their proposed methodology.
Open Datasets Yes Victim Models and Datasets. We set up six victim models (see column FV in Table 1), each model trained on a popular image classification dataset.
Dataset Splits No We train and evaluate each victim model on their respective train and test sets.
Hardware Specification Yes The reported numbers were summarized over 10K unique predictions performed on an Nvidia Tesla V100.
Software Dependencies No The paper does not specify software dependencies with version numbers.
Experiment Setup Yes All models are trained using SGD (LR = 0.1) with momentum (0.5) for 30 (Le Net) or 100 epochs (VGG16), with a LR decay of 0.1 performed every 50 epochs.