Defending against Model Stealing via Verifying Embedded External Features

Authors: Yiming Li, Linghui Zhu, Xiaojun Jia, Yong Jiang, Shu-Tao Xia, Xiaochun Cao1464-1472

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

Reproducibility Variable Result LLM Response
Research Type Experimental We examine our method on both CIFAR-10 and Image Net datasets. Experimental results demonstrate that our method is effective in detecting different types of model stealing simultaneously, even if the stolen model is obtained via a multi-stage stealing process.
Researcher Affiliation Academia 1Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, China 2Research Center of Artificial Intelligence, Peng Cheng Laboratory, Shenzhen, China 3Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes The codes for reproducing main results are available at Github (https://github.com/zlh-thu/Stealing Verification).
Open Datasets Yes We evaluate our defense on CIFAR-10 (Krizhevsky, Hinton et al. 2009) and (a subset of) Image Net (Deng et al. 2009) dataset.
Dataset Splits No The paper does not explicitly provide information about validation dataset splits for reproducing the main experiments.
Hardware Specification No The paper does not provide any specific hardware specifications (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies No The paper does not provide specific software dependencies or library versions with numbers (e.g., Python 3.8, PyTorch 1.9) needed to replicate the experiment.
Experiment Setup Yes We poison 10% training samples for all defenses. Besides, we adopt a white-square as the trigger pattern for Bad Nets and adopt a oil paint as the style image for our defense.