Event № 1338

 We present an algorithmic framework for linear and nonlinear expe rimental design obtained by the minimization of an empirical estimate of th=rimental design obtained by the minimization of an empirical estimate of the risk. The formulation leads to a bilevel stochastic optimization problem that is challenging to solve.  We discuss the solution of the problem and show that for linear inverse problems this can be done efficiently. We also discuss techniques for the nonlinear case that are based on stochastic programming techniques.  Finally, we demonstrate that design in inverse problems can yield significant advantages and reduce the risk as well as the cost of experiments.                        

NOTE UNUSUAL ROOM, DAY AND TIME