CLPA: Clean-Label Poisoning Availability Attacks Using Generative Adversarial Nets

Authors: Bingyin Zhao, Yingjie Lao9162-9170

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the effectiveness of our approach on CIFAR-10 and Image Net dataset over a variety type of models.
Researcher Affiliation Academia Department of Electrical and Computer Engineering, Clemson University, SC, 29634, USA {bingyiz, ylao}@clemson.edu
Pseudocode Yes Algorithm 1: Poisoned Data Generation and Algorithm 2: Negative Class Selection
Open Source Code Yes Codes are available at: https://github.com/bxz9200/CLPA.
Open Datasets Yes We demonstrate the effectiveness of our approach on CIFAR-10 and Image Net dataset
Dataset Splits No The paper states: "we train the downstream classifiers using 5000 training images from the training dataset combined with poisoned data" and "The performance is evaluated on the test dataset of 10000 images." It does not explicitly mention a separate validation dataset split.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory amounts used for running the experiments.
Software Dependencies No The paper mentions software components and models (e.g., "conditional GAN", "Big GAN", "SGD optimizer", "Inception-V3 model") but does not provide specific version numbers for any of them.
Experiment Setup Yes The neural networks are trained 10 epochs for end-to-end training and FC layer only. ... SGD optimizer is used for model training at a learning rate of 1 10 4 with a batch size of 16." (from section "Evaluation on CIFAR-10") and "Training iterations of phase I", "Training iterations of phase II", "Margin parameter α" in Table 2.