Reproduce “Calibrated Multiaccuracy: Deterministic and Randomized Sample Complexity” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiuyao Lu, Georgy Noarov, Aaron Roth
The authors characterize minimax sample complexity for calibrated multiaccuracy: overall calibration plus small average prediction-error correlation across a test family. Up to polylogarithmic factors, randomized predictors need ε⁻⁵ᐟ² samples when A is of order log(1/ε), versus ε⁻⁸ᐟ³ for deterministic predictors. The deterministic lower bound holds only when A ranges from order log(1/ε) to order ε⁻¹; A is the log of the test-family size.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.