Faster Algorithms and Constant Lower Bounds for the Worst-Case Expected Error

Neural Information Processing Systems 

"snowball sampling"), we show that our