Mingda Qiao

We study the tradeoff between sample complexity and round complexity in on-demand sampling.

We introduce a model of prediction with limited selectivity, and prove instance-dependent bounds on the optimal error rate.