Diameter-based Interactive Structure Search.
Christopher Tosh, Daniel Hsu · arXiv (Cornell University) · 2019
In this work, we introduce \emph{interactive structure search}, a generic framework that encompasses many interactive learning settings, both explored and unexplored. We show that a recently developed active learning algorithm of~\citet{TD17} can be adapted for interactive structure search, that it can be made noise-tolerant, and that it enjoys favorable convergence rates.