B0747
Title: Langevin-type sampling algorithm for non-log-concave non-smooth distributions
Authors: Shogo Nakakita - The University of Tokyo (Japan) [presenting]
Abstract: An approximate sampling algorithm is considered for a distribution whose density function is neither log-concave nor smooth. The proposed algorithm combines the unadjusted Langevin algorithm with empirical mollification to approximate the smoothed weak gradient of the potential function. Under a dissipativity condition on the potential function and a stability condition on its weak gradient, the complexity to let the 2-Wasserstein distance between the distribution of the algorithm and the target distribution is analysed arbitrarily small.
