Bayesian Differential Privacy for Machine Learning

Authors: Aleksei Triastcyn, Boi Faltings

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

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments demonstrate that in-distribution samples in classic machine learning datasets, such as MNIST and CIFAR-10, enjoy significantly stronger privacy guarantees than postulated by DP, while models maintain high classification accuracy. This experimental section comprises two parts. First, we examine how well Bayesian DP composes over multiple steps. ... Second, we consider the context of machine learning. In particular, we use the differentially private stochastic gradient descent (DP-SGD), ... to train neural networks on classic image classification tasks MNIST (Le Cun et al., 1998) and CIFAR10 (Krizhevsky, 2009).
Researcher Affiliation Academia 1Artificial Intelligence Lab, Ecole Polytechnique F ed erale de Lausanne (EPFL), Lausanne, Switzerland.
Pseudocode No The paper describes its methods mathematically and in text, but does not include structured pseudocode or algorithm blocks.
Open Source Code Yes For more details and additional experiments, we refer the reader to the supplementary material, while the source code is available on Git Hub2. 2https://github.com/Aleksei Triastcyn/ bayesian-differential-privacy
Open Datasets Yes train neural networks on classic image classification tasks MNIST (Le Cun et al., 1998) and CIFAR10 (Krizhevsky, 2009). We also perform experiments with variational inference on Abalone (Waugh, 1995) and Adult (Kohavi, 1996) datasets.
Dataset Splits No The paper mentions training and test sets and discusses epochs, but does not provide specific details on training/validation/test splits, such as percentages, sample counts, or cross-validation setup.
Hardware Specification No The paper does not provide specific details about the hardware used for experiments, such as CPU/GPU models or memory specifications.
Software Dependencies No The paper mentions general techniques and frameworks like 'DP-SGD' and 'variational inference', but does not specify any software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, TensorFlow 2.x).
Experiment Setup No The paper describes the general approach of clipping gradient norms and adding Gaussian noise in DP-SGD, and mentions '50 epochs' for MNIST, but it does not provide specific hyperparameter values like learning rates, batch sizes, or exact clipping and noise parameters (C and σ) for its experiments.