Differentially Private Domain Adaptation with Theoretical Guarantees

Authors: Raef Bassily, Corinna Cortes, Anqi Mao, Mehryar Mohri

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental While our main objective is a theoretical analysis, we also report the results of several experiments. We first show that the non-private versions of our algorithms match state-of-the-art performance in supervised adaptation and that for larger values of the target sample size or ε, the performance of our private algorithms remains close to that of their non-private counterparts.
Researcher Affiliation Collaboration Raef Bassily 1 Corinna Cortes 2 Anqi Mao 3 Mehryar Mohri 2 3 1The Ohio State University 2Google Research, New York, NY; 3Courant Institute of Mathematical Sciences, New York, NY. Correspondence to: Anqi Mao <aqmao@cims.nyu.edu>.
Pseudocode Yes Algorithm 1 Cnvx Adap Private adaptation algorithm based on F
Open Source Code No The paper does not contain any explicit statement about releasing source code or a link to a code repository.
Open Datasets Yes We consider five regression datasets with dimensions as high as 384 from the UCI machine learning repository (Dua & Graff, 2017), the Wind, Airline, Gas, News and Slice.
Dataset Splits Yes We carry out model selection on the target validation set and report in Table 1 the mean and standard deviation on the test set over 10 random splits of the target training and validation sets.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models or processor types used for running its experiments.
Software Dependencies No The paper mentions using logistic regression classifiers but does not specify any software dependencies with version numbers.
Experiment Setup Yes For details on hyperparameter tuning see Appendix E.