Learning-augmented private algorithms for multiple quantile release

Authors: Mikhail Khodak, Kareem Amin, Travis Dick, Sergei Vassilvitskii

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conclude with experiments on challenging tasks demonstrating that learning predictions across one or more instances can lead to large error reductions while preserving privacy.
Researcher Affiliation Collaboration 1Carnegie Mellon University; work done in part as an intern at Google Research New York. 2Google Research New York.
Pseudocode Yes Algorithm 2: Approximate Quantiles with predictions
Open Source Code Yes Code to reproduce our results is available at https://github.com/mkhodak/private-quantiles.
Open Datasets Yes We evaluate this approach... on Adult (Kohavi, 1996) and Goodreads (Wan & Mc Auley, 2018)...
Dataset Splits Yes Adult tests the D D1 case, with its train set the public dataset and a hundred samples from test as private.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as exact GPU or CPU models, memory configurations, or specific cloud computing instance types.
Software Dependencies No The paper mentions various software components and packages used, such as COCOB (with a GitHub link), textstat (with a GitHub link), NLTK, and an adaptation of DP-FTRL from Google Research. However, it does not provide specific version numbers for any of these dependencies or the general programming environment (e.g., Python version, PyTorch/TensorFlow versions).
Experiment Setup Yes We use the following reasonable guesses for locations ν, scales σ, and quantile ranges ra, bs for these distributions: age: ν 40, σ 5, a 10, b 120; hours: ν 40, σ 2, a 0, b 168; rating: ν 2.5, σ 0.5, a 0, b 5; page count: ν 200, σ 25, a 0, b 1000