Speeding up convergence to equilibrium for diffusion processes Seminar

In many applications it is necessary to sample from a probability distribution in high dimensional spaces. A well known technique for doing this is running appropriate stochastic dynamics that is ergodic with respect to the distribution form which we want to sample. For the standard (reversible) over damped Langevin dynamics convergence can be quite slow when the target distribution is multimodal. In this talk we present techniques for accelerating convergence to equilibrium by modifying the dynamics through the addition of a drift that breaks the reversibility (detailed balance) of the dynamics.