WaNet - Imperceptible Warping-based Backdoor Attack

Authors: Tuan Anh Nguyen, Anh Tuan Tran

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our attack method achieves invisibility without sacrificing accuracy. It performs similarly to stateof-the-art backdoor methods in terms of clean and attack accuracy, verified on common benchmarks such as MNIST, CIFAR-10, GTSRB, and Celeb A. Our attack is also undetectable by various backdoor defense mechanisms; none of existing algorithms can recognize or mitigate our backdoor.
Researcher Affiliation Collaboration 1Vin AI Research, 2Hanoi University of Science and Technology, 3Vin University
Pseudocode No The paper describes the process of their method using text and figures but does not include formal pseudocode or algorithm blocks.
Open Source Code Yes Our code is publicly available at https://github.com/Vin AIResearch/ Warping-based_Backdoor_Attack-release.
Open Datasets Yes Following the previous backdoor attack papers, we performed experiments on four datasets: MNIST (Le Cun et al., 1998), CIFAR-10 (Krizhevsky et al., 2009), GTSRB (Stallkamp et al., 2012) and Celeb A (Liu et al., 2015).
Dataset Splits Yes MNIST... This dataset consists of 70,000 grayscale, 28 28 images, divided into a training set of 60,000 images and a test set of 10,000 images. CIFAR-10... is divided into two subsets: a training set of 50,000 images and a test set of 10,000 images. GTSRB... It is divided into a training set of 39,209 images and a test set of 12,630.
Hardware Specification Yes We use a system of a GPU RTX 2080Ti and a CPU i7 9700K to conduct our experiment.
Software Dependencies No The paper mentions PyTorch for implementing W, but it does not specify a version number or list any other software dependencies with version numbers.
Experiment Setup Yes The initial learning rate was 0.01, which was reduced by a factor of 10 after each 100 training epochs. The networks were trained until convergence. We used k = 4, s = 0.5, ρa = 0.1, and ρn = 0.2.