Generalized Linear Bandits with Local Differential Privacy

Authors: Yuxuan Han, Zhipeng Liang, Yang Wang, Jiheng Zhang

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we conduct experiments with both simulation and real-world datasets to demonstrate the consistently superb performance of our algorithms under LDP constraints with reasonably small parameters (ε, δ) to ensure strong privacy protection.
Researcher Affiliation Academia The Hong Kong University of Science and Technology
Pseudocode Yes Algorithm 1: LDP Single-parameter Contextual Bandit and Algorithm 2: LDP Multi-parameter Contextual Bandit
Open Source Code Yes The source code to reproduce all the results is available at the Git Hub repo liangzp/LDP-Bandit.
Open Datasets Yes On-Line Auto Lending dataset CRPM-12-001 provided by Columbia University https://www8.gsb. columbia.edu/cprm/research/datasets, and has been used in the study of contextual bandits by [26, 11].
Dataset Splits No The paper does not explicitly provide specific percentages or counts for training, validation, or test splits. It mentions 'synthetic datasets' and 'real-world datasets' without specifying how they were partitioned.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, memory, or cloud instance types used for running its experiments. It only mentions 'simulation' and 'real-world datasets' without hardware context.
Software Dependencies No The paper does not provide specific software dependencies with version numbers. It refers to 'scikit-learn' through citation [25], but no version is specified. It also mentions 'LDP-UCB' and 'LDP-GLOC' as methods being compared, but these are not software dependencies in the sense of libraries with versions.
Experiment Setup Yes We evaluate all the four methods on two different privacy levels ε = 0.5 and 1 in synthetic datasets, which are industry standards. For the sake of comparison, the learning step parameter for LDP-GLOC and LDP-SGD are tuned in the same way.