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 and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Private Evolution Converges
Authors: Tomás González Lara, Giulia Fanti, Aaditya Ramdas
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work, we develop a new theoretical framework to understand PE s practical behavior and identify sufficient conditions for its convergence. ... We demonstrate the practical relevance of our theoretical findings in experiments. ... 6 Empirical Results |
| Researcher Affiliation | Academia | Tomas Gonzalez Carnegie Mellon University EMAIL Giulia Fanti Carnegie Mellon University EMAIL Aaditya Ramdas Carnegie Mellon University EMAIL |
| Pseudocode | Yes | Algorithm 1 (Theoretical) Private Evolution; steps in blue differ from practical PE [33] ... Algorithm 2 Private Evolution with DBL projection; steps in blue differ from practical PE [33] ... Algorithm 3 (Practical) Private Evolution [33] ... Algorithm 4 Nearest Neighbor Histogram (NN_histogram) ... Algorithm 5 Private signed measured mechanism with approx DP |
| Open Source Code | Yes | We will provide the code in the Supplemental Material. |
| Open Datasets | Yes | Next, we evaluate our theoretical predictions on an image modeling task, using two subclasses (Dog and Plane) of images from the CIFAR-10 dataset [30]. |
| Dataset Splits | No | The paper mentions using 'CIFAR-10 dataset' but does not explicitly provide information about how the dataset was split into training, testing, or validation sets in the main text. It only states 'using two subclasses (Dog and Plane) of images from the CIFAR-10 dataset'. |
| Hardware Specification | Yes | Simulations are run on a personal laptop with Apple M2 Pro chip and 32GB of memory. |
| Software Dependencies | No | The paper does not explicitly list specific software components with version numbers required to replicate the experiment. |
| Experiment Setup | Yes | We use privacy parameters ε = 1, δ = 10 4. Our theory indicates that O(log(nε)) PE steps ensure convergence; for simulations we set T = 2 log(nε). The noise σ is then computed with the analytic Gaussian mechanism [4, Theorem 8]. The remaining parameters are set as in Theorem 4.1. ... We set the privacy parameters to ε = 5, δ = 10 4. For each n {100, 200, ..., 600}, we set the hyperparameters according to Theorem 4.1 and run PE. |