Graphical-model based estimation and inference for differential privacy

Authors: Ryan Mckenna, Daniel Sheldon, Gerome Miklau

ICML 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we measure the accuracy and scalability improvements enabled by probabilistic graphical-model (PGM) based estimation when it is incorporated into existing privacy mechanisms.
Researcher Affiliation Academia 1University of Massachusetts, Amherst 2Mount Holyoke College.
Pseudocode Yes Algorithm 1 Proximal Estimation Algorithm; Algorithm 2 Accelerated Proximal Estimation Algorithm
Open Source Code No The paper does not include any statement about making its source code publicly available or provide a link to a code repository for the methodology described.
Open Datasets No The paper uses datasets such as 'Titanic', 'Adult', 'Loans', and 'Stroke' and lists their properties in Table 1. However, it does not provide direct links, DOIs, repository names, or specific bibliographic citations for accessing these datasets.
Dataset Splits No The paper focuses on estimating data distributions from noisy measurements and evaluating query accuracy, rather than training a predictive model using traditional train/validation/test dataset splits. Therefore, no such explicit splits are mentioned or provided for the experimental setup.
Hardware Specification No The paper states: 'Experiments are done on 2 cores of a single compute cluster node with 16 GB of RAM and 2.4 GHz processors.' This provides some general specifications but lacks specific CPU or GPU models, or more detailed hardware components needed for full reproducibility.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., Python 3.8, PyTorch 1.9). While it mentions tools like 'Autograd' and 'LSMR', it does not list their versions or other required software with specific versions.
Experiment Setup Yes We run Algorithm 1 with line search for Dual Query and Algorithm 2 for the other mechanisms, each for 10000 iterations. We use a privacy budget of = 1.0 (and δ = 0.001 for Dual Query). The variable ηt in this algorithm is a step size, which can be constant, decreasing, or found via line search.