Skip to main content
Two methods for assumption-light inference
Add to Calendar 2021-09-02T19:30:00 2021-09-02T20:30:00 UTC Two methods for assumption-light inference Zoom
Start DateThu, Sep 02, 2021
3:30 PM
End DateThu, Sep 02, 2021
4:30 PM
Presented By
Sivaraman Balakrishnan (Carnegie Mellon University)
Event Series: Statistics Colloquia


In this talk, I will discuss two methods -- universal inference and the HulC -- which use sample-splitting to construct confidence sets which are valid under very weak regularity conditions. Universal inference uses sample-splitting to ease the construction of likelihood-ratio based confidence sets. The HulC uses sample-splitting to significantly weaken the regularity conditions needed for the validity of classical resampling based methods (the bootstrap and sub-sampling). These methods are easy to apply, yield valid inference in a wide-range of challenging problems, and achieve these strong guarantees at a surprisingly small statistical price.

The two main papers I will discuss are:

This is based on joint work with Arun Kuchibhotla, Aaditya Ramdas and Larry Wasserman.

To learn more about Sivaraman visit: