Generating Private Synthetic Data with Genetic Algorithms

Authors: Terrance Liu, Jingwu Tang, Giuseppe Vietri, Steven Wu

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate empirically that on data with both discrete and real-valued attributes, PRIVATE-GSD outperforms the state-of-the-art methods on nondifferential queries while matching accuracy in approximating differentiable ones.
Researcher Affiliation Academia 1Carnegie Mellon University 2Peking University 3University of Minnesota.
Pseudocode Yes Algorithm 1 Private Genetic Algorithm for Synthetic Data (PRIVATE-GSD)
Open Source Code Yes The PRIVATE-GSD source code is publicly available at https: //github.com/giusevtr/private_gsd.
Open Datasets Yes For our empirical evaluation, we use datasets derived from the Folktables package (Ding et al., 2021), which defines datasets using samples from the American Community Survey (ACS).
Dataset Splits No For the main experiments, the paper mentions using datasets from the Folktables package but does not specify any training/validation/test splits. For the ML evaluation in Appendix C, it states 'dividing each dataset into a training and test set, using an 80/20 partition,' but no separate validation split is mentioned.
Hardware Specification No The paper does not explicitly describe the specific hardware (e.g., GPU/CPU models, memory specifications) used for running its experiments.
Software Dependencies No The paper mentions using the 'Folktables package (Ding et al., 2021)' but does not provide specific version numbers for any software dependencies, libraries, or frameworks used in the experiments.
Experiment Setup Yes Table 7. Hyperparameters experiments (with adaptivity). lists detailed hyperparameter values such as Data Size (N), Pmut, Pcross, Elite Size, Max Generations, Queries Sampled (K), Learning Rate, Inverse Temp. (σit), # Product Mixtures (K), Batch Size (B), Max Iterations (M), # Samples, and T (adaptive epochs) for various methods.