Guaranteed validity for empirical approaches to adaptive data analysis

Files
rogers20a.pdf(635.06 KB)
Published version
Date
2020-08-15
DOI
Authors
Rogers, Ryan
Roth, Aaron
Smith, Adam
Srebro, Nathan
Thakkar, Om
Woodworth, Blake
Version
OA Version
Citation
R. Rogers, A. Roth, A. Smith, N. Srebro, O. Thakkar, B. Woodworth. 2020. "Guaranteed Validity for Empirical Approaches to Adaptive Data Analysis." 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020
Abstract
We design a general framework for answering adaptive statistical queries that focuses on providing explicit confidence intervals along with point estimates. Prior work in this area has either focused on providing tight confidence intervals for specific analyses, or providing general worst-case bounds for point estimates. Unfortunately, as we observe, these worst-case bounds are loose in many settings — often not even beating simple baselines like sample splitting. Our main contribution is to design a framework for providing valid, instance-specific confidence intervals for point estimates that can be generated by heuristics. When paired with good heuristics, this method gives guarantees that are orders of magnitude better than the best worst-case bounds. We provide a Python library implementing our method.
Description
License
Copyright 2020 by the author(s).