Sayak Chakrabarty

We establish the first parameter-free CLT for m-out-of-n bootstrap quantiles, sharpen it with Edgeworth bounds for finite samples, and show gains for different learning tasks.