Mitigating Privacy Risk in Membership Inference by Convex-Concave Loss

Authors: Zhenlong Liu, Lei Feng, Huiping Zhuang, Xiaofeng Cao, Hongxin Wei

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we validate the effectiveness of our CCL across a wide range of datasets with diverse models, various attack models, and multiple defense baselines.
Researcher Affiliation Academia Zhenlong Liu 1 Lei Feng 2 Huiping Zhuang 3 Xiaofeng Cao 4 Hongxin Wei 1 1Southern University of Science and Technology 2Singapore University of Technology and Design 3South China University of Technology 4Jilin University.
Pseudocode No The paper describes the proposed method in prose and mathematical equations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Our code is available at https://github.com/ml-stat-Sustech/Convex Concave Loss.
Open Datasets Yes In our evaluation, we employ five datasets: Texas100 (Texas Department of State Health Services, 2006), Purchase100 (Kaggle, 2014), CIFAR-10, CIFAR100 (Krizhevsky et al., 2009), and Image Net (Russakovsky et al., 2015).
Dataset Splits Yes For standard training methods, we split each dataset into four subsets, with each subset serving alternately as the training or testing set for the target and shadow models.
Hardware Specification No The paper does not provide specific details regarding the hardware (e.g., GPU models, CPU types, or cloud computing instances) used for running the experiments.
Software Dependencies No The paper does not specify version numbers for any software dependencies or libraries used in the implementation or experimentation.
Experiment Setup Yes We train the models using SGD with a momentum of 0.9, a weight decay of 0.0005, and a batch size of 128. We set the initial learning rate as 0.1 and drop it by a factor of 10 at each decay epoch.