White-box vs Black-box: Bayes Optimal Strategies for Membership Inference
Authors: Alexandre Sablayrolles, Matthijs Douze, Cordelia Schmid, Yann Ollivier, Herve Jegou
ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section we evaluate the membership inference methods on machine-learning tasks of increasing complexity. Table 1 shows the results of our experiments, in terms of accuracy and mean average precision. |
| Researcher Affiliation | Collaboration | 1University Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK 2Facebook AI Research. |
| Pseudocode | No | The paper describes methods conceptually and mathematically, but does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | No | The paper does not contain any explicit statements about releasing source code or provide links to a code repository for the methodology described. |
| Open Datasets | Yes | CIFAR-10 is a dataset of 32 × 32 pixel images grouped in 10 classes. We evaluate a real-world dataset and tackle classification with large neural networks on the Imagenet dataset (Deng et al., 2009; Russakovsky et al., 2015), which contains 1.2 million images partitioned into 1000 categories. |
| Dataset Splits | Yes | We assume that λ = 1/2 (n/2 training images and n/2 test images). We also reserve n/2 images to estimate 0 for the MATT method. The regularization parameter C of the logistic regression is cross-validated on held-out data. |
| Hardware Specification | No | The paper does not provide any specific hardware details such as GPU models, CPU types, or memory used for running the experiments. |
| Software Dependencies | No | The paper mentions machine learning models and techniques (e.g., 'logistic regression', 'Resnet18'), but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions). |
| Experiment Setup | Yes | Our model is trained for 50 epochs with a learning rate of 0.001. The model is learned in 90 epochs, with an initial learning rate of 0.01, divided by 10 every 30 epochs. Parameter optimization is conducted with SGD with a momentum of 0.9, a weight decay of 10^-4, and a batch size of 256. |