Estimating conditional distributions in high dimension.

In many applications one is interested in computing conditional distributions of stochastic models given observed data. Exact computations are rarely possible, and
approximation techniques adopting Monte Carlo samples are widely used for their nice statistical properties. However, in many situations the approximation error of Monte Carlo
algorithms grows exponentially with the model dimension. In this talk I will discuss the main ingredients to develop local Monte Carlo algorithms that can avoid the curse of
dimensionality by yielding dimension-free approximation errors. (Joint work with R. van Handel)