Label Robust and Differentially Private Linear Regression: Computational and Statistical Efficiency

Authors: Xiyang Liu, Prateek Jain, Weihao Kong, Sewoong Oh, Arun Suggala

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

Reproducibility Variable Result LLM Response
Research Type Experimental We present experimental results comparing our proposed technique (DP-ROBGD) with other baselines. We consider non-corrupted regression in this section and defer corrupted regression to the App. K. We begin by describing the problem setup and the baseline algorithms first.
Researcher Affiliation Collaboration Xiyang Liu Paul Allen School of Computer Science & Engineering University of Washington xiyangl@cs.washington.edu Prateek Jain Google Research prajain@google.com Weihao Kong Google Research weihaokong@google.com Sewoong Oh Paul Allen School of Computer Science & Engineering University of Washington, and Google Research sewoong@cs.washington.edu Arun Sai Suggala Google Research arunss@google.com
Pseudocode Yes Algorithm 1: Robust and Private Linear Regression
Open Source Code No The paper does not provide an explicit statement about releasing source code for the described methodology, nor does it provide a direct link to a code repository.
Open Datasets No The paper states, 'We generate data for all the experiments using the following generative model.' This indicates synthetic data generation rather than the use of a publicly available dataset with concrete access information.
Dataset Splits No The paper mentions using 20% of samples for an internal calculation ('to compute γt in line 5') but does not specify standard training, validation, and test dataset splits needed for reproducibility. It primarily uses synthetically generated data without predefined splits.
Hardware Specification No The paper does not specify any hardware details (e.g., GPU/CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not list specific software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x).
Experiment Setup Yes We set the DP parameters (ϵ, δ) as ϵ = 1, δ = min(10 6, n 2)... Instead of relying on Private Norm Estimator to estimate Γ, we set it to its true value Tr(Σ). This is done for a fair comparison with DP-AMBSSGD which assumes the knowledge of Tr(Σ). Next, we use 20% of the samples to compute γt in line 5 (instead of the 50% stated in Alg. 1). In our experiments we also present results for a variant of our algorithm called DP-ROBGD* which outputs the best iterate based on γt, instead of the last iterate.