Differentially Private Maximal Information Coefficients

Authors: John Lazarsfeld, Aaron Johnson, Emmanuel Adeniran

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We perform experiments on a variety of real and synthetic datasets.
Researcher Affiliation Collaboration 1Department of Computer Science, Yale University 2U.S. Naval Research Laboratory. Correspondence to: John Lazarsfeld <john.lazarsfeld@yale.edu>.
Pseudocode No The paper describes its mechanisms (e.g., Mechanism 1, Mechanism 2, Mechanism 3) in prose, and refers to the OPTIMIZEAXIS routine, but it does not include any structured pseudocode or algorithm blocks.
Open Source Code Yes The code and data used to obtain our experimental results can be accessed at https://github.com/jlazarsfeld/dp-mic, and more implementation details are given in Appendix E.1.
Open Datasets Yes We used synthetic data (following the methodology of Reshef et al. (2011; 2016))...The Spellman data (Spellman et al., 1998; Reshef et al., 2011)...Finally, the Baseball dataset (Reshef et al., 2011; Prospectus, Accessed March 2020)...The code and data used to obtain our experimental results can be accessed at https://github.com/jlazarsfeld/dp-mic
Dataset Splits No The paper describes parameter tuning on synthetic data and analyzes bias/variance, which functionally serves as validation. However, it does not explicitly provide details about 'training/test/validation dataset splits' using standard terminology or specific proportions for model training.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions using 'the MINEPY library implementation from Albanese et al. (2018)' but does not provide a specific version number for MINEPY or any other software dependencies.
Experiment Setup Yes We set its parameters to B(n) = n0.6 and c = 15...We considered sample sizes of n {25, 250, 500, 1000, 5000, 10000}, ϵ {0.1, 1.0}, and various values of B between 4 and 150...we fixed c = 5 for the MICr-Lap mechanism...and for the MICr-Geom mechanism we considered c {1, 2}...