5th Kalman Lecture by Nicolas Chopin

Nested cubing integration: how to get a O(N{-10}) error when you compute your favourite integral

by Nicolas Chopin (ENSAE, Institut Polytechnique de Paris, France)

Please find the record here.

This talk will explain why computing integrals remains such an important task in applied mathematics, in particular in Statistics and machine learning. I will present several known methods to get unbiased estimates of an integral, and an old result giving the best achievable error rate given the number of evaluations and the regularity of the function. I will then present a new method, developed with Mathieu Gerber, which attains this rate. The underlying ideas of this method are surprisingly simple, and rely on concepts covered in undergraduate courses.

- joint work with Mathieu Gerber, Bristol University, UK