Understanding Backdoor Attacks through the Adaptability Hypothesis

Authors: Xun Xian, Ganghua Wang, Jayanth Srinivasa, Ashish Kundu, Xuan Bi, Mingyi Hong, Jie Ding

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on benchmark image datasets and state-of-the-art backdoor attacks for deep neural networks are conducted to corroborate the hypothesis. Theoretical analyses of backdoor attacks under classical machine learning context.
Researcher Affiliation Collaboration 1Department of ECE, University of Minnesota 2School of Statistics, University of Minnesota 3Cisco Research 4Carlson School of Management, University of Minnesota.
Pseudocode Yes G. Pseudo-code for visualisation algorithms and additional experimental results. Algorithm 1 Visualizing high-dimensional data
Open Source Code No The paper does not contain any explicit statement about open-sourcing its code or a link to a code repository.
Open Datasets Yes We use 3 popular datasets: MNIST (Le Cun et al., 2010), CIFAR10 (Krizhevsky et al., 2009), and GTSRB (Stallkamp et al., 2012).
Dataset Splits No The paper mentions training and test sets but does not provide specific details about a validation split, explicit percentages for data partitioning, or cross-validation setup.
Hardware Specification Yes All of our experiments are conducted on a workstation with one A100 GPU.
Software Dependencies No The paper mentions machine learning models (Le Net, Res Net, VGG) and optimization algorithms (SGD) but does not specify software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes For Res Net and VGG models, we adopt the standard training pipeline of SGD with a momentum of 0.9, a weight decay of 10-4, and a batch size of 128 for optimization. For Let Net, we adopt the standard training pipeline of SGD with the initial learning rate of 0.1/0.01.