Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Smoothed Differential Privacy
Authors: Ao Liu, Yu-Xiang Wang, Lirong Xia
TMLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimentally, we verify that, according to smoothed DP, the discrete sampling mechanisms are private in real-world elections, and some discrete neural networks can be private without adding any additive noise. |
| Researcher Affiliation | Collaboration | Ao Liu EMAIL Core Machine Learning, Google Yu-Xiang Wang EMAIL Department of Computer Science, UC Santa Barbara Lirong Xia EMAIL Computer Science Department, Rensselaer Polytechnic Institute |
| Pseudocode | Yes | Algorithm 1: Calculate the (exact) privacy profile δ for smoothed DP Algorithm 2: Sampling-histogram mechanism MH Algorithm 3: Continuous sampling-average mechanism MA |
| Open Source Code | No | The paper does not provide explicit links to source code repositories, an explicit statement of code release for their methodology, or indicate code in supplementary materials. While it mentions MATLAB for implementation, it doesn't provide the code developed by the authors. |
| Open Datasets | Yes | Experimentally, we numerically evaluate the privacy level of the sampling-histogram mechanism using US presidential election data. ... the 2020 presidential election. ... Res Net-18 network trained on CIFAR-10 database (Banner et al., 2018). |
| Dataset Splits | No | The paper describes sampling T = η n data without replacement (e.g., 'batch size T = η n'). However, it does not specify explicit training, validation, and testing dataset splits for evaluation. The election data uses distributions, and the SGD experiment mentions batch size but not overall dataset splits for evaluation. |
| Hardware Specification | Yes | All experiments of this paper are implemented in MATLAB 2021a and tested on a Windows 10 Desktop with an Intel Core i7-8700 CPU and 32GB RAM. |
| Software Dependencies | Yes | All experiments of this paper are implemented in MATLAB 2021a and tested on a Windows 10 Desktop with an Intel Core i7-8700 CPU and 32GB RAM. |
| Experiment Setup | Yes | We use a similar setting as the motivating example, where 0.2% of the votes are randomly lost. ... We thus let the set of distributions Π = {N8-bit(0, 0.122), N8-bit(0.2, 0.122)}, where N8-bit(µ, σ2) denotes the 8-bit quantized Gaussian distribution ... The standard variation, 0.12, is the same as the standard variation of gradients in a Res Net-18 network trained on CIFAR-10 database (Banner et al., 2018). We use the standard setting of batch size T = n. ... We find that δ is also exponentially small when ϵ = 0.5, 1 or 2 |