Hidden Trigger Backdoor Attacks

Authors: Aniruddha Saha, Akshayvarun Subramanya, Hamed Pirsiavash11957-11965

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

Reproducibility Variable Result LLM Response
Research Type Experimental We perform an extensive study on various image classification settings and show that our attack can fool the model by pasting the trigger at random locations on unseen images although the model performs well on clean data. We also show that our proposed attack cannot be easily defended using a state-of-the-art defense algorithm for backdoor attacks. [...] 4. Experiments
Researcher Affiliation Academia Aniruddha Saha, Akshayvarun Subramanya, Hamed Pirsiavash University of Maryland, Baltimore County {anisaha1, akshayv1, hpirsiav}@umbc.edu
Pseudocode Yes Algorithm 1: Generating poisoning data
Open Source Code No The paper does not provide an explicit statement about open-sourcing the code for the methodology described, nor does it include a link to a code repository.
Open Datasets Yes We divide the Image Net data to three sets for each category: 200 images for generating the poisoned data, 800 images for training the binary classifier, and 100 images for testing the binary classifier. [...] We also use CIFAR10 dataset for the experiments in Section 4.4
Dataset Splits Yes Dataset: Since we want to have separate datasets for generating poisoned data and finetuning the binary model, we divide the Image Net data to three sets for each category: 200 images for generating the poisoned data, 800 images for training the binary classifier, and 100 images for testing the binary classifier. [...] For each category, we have 1,500 images to train the poisoned data, 1,500 images for finetuning, and 1,000 images for evaluation. These three sets are disjoint.
Hardware Specification Yes It takes about 5 minutes to generate 100 poisoned images on a single NVIDIA Titan X GPU.
Software Dependencies No The paper mentions using Alex Net and the PGD attack, but does not provide specific version numbers for any software libraries, frameworks, or programming languages (e.g., PyTorch, TensorFlow, Python version).
Experiment Setup Yes For our Image Net experiments we set a reference parameter set where the perturbation ϵ = 16, trigger size is 30x30 (while images are 224x224), and we randomly choose a location to paste the trigger on the source image. We generate 100 poisoned examples and add to our target class training set of size 800 images during finetuning. Thus about 12.5% of the target data is poisoned. To generate our poisoned images, we run Algorithm 1 with mini-batch gradient descent for 5,000 iterations with a batch size of K = 100. We use an initial learning rate of 0.01 with a decay schedule parameter of 0.95 every 2,000 iterations.