This past Tuesday I gave a talk in the Natural Sciences Seminar Series at Pepperdine. The talk (which has now been posted on my Research talk web page here) is another in the series about TPA.
Abstract. Monte Carlo methods use random variates in order to approximate high dimensional integrals. The resulting estimates have a variance that can vary widely from problem to problem. If constructed poorly the resulting variance can even be infinite. In this talk I will present a new method for estimating integrals where the tightness of the estimate is a direct function of the number of samples. In other words, there is no need to worry about the variance of the estimate, it can be set to any level of precision that the user desires. The method operates by drawing random variates from a sequence of distributions that is determined adaptively at the running time of the algorithm.