RADU HERBEI

My current research interests include Markov chain Monte Carlo and Bayesian methods, and their applications to inverse problems, in particular to atmospheric sciences. Inverse problems are usually ill- posed, hence a solution may not exist, or, there may be multiple solutions. In the latter case, robust solutions can be obtained by modeling the observation error and introducing prior information on the parameters. MCMC methods are then used to sample the posterior distribution. This requires solving the "forward problem" as well, for a given set of parameters. Relevant issues here include prior specification, additional model constraints to identify "the solution", fast / parallel computation, multi-scale methods. I am also studying convergence rates of the MCMC sampler under this setup. Additionally, I am working on multi-class classification problems with a reject option.