We are developing advanced optimization algorithms to tackle unsolved problems that arise in science applications. Our main focus is on the development of efficient and effective algorithms for optimization problems whose objective functions are obtained by running computationally expensive black box simulations and for which derivative information is not available (see the illustration below).
We have solution methods for problems with
- continuous, integer, and mixed-integer parameters
- computationally cheap and expensive black box constraints
- multiple conflicting black box objective functions
- inexpensive objective functions and expensive black box constraints
We are also open to other problem classes (fast to compute functions, given derivatives, etc) and we would be happy to give advice on useful methods for your specific problem!
General optimization workflow: With a given set of parameters, x, we run the expensive simulation, derive a cost function value, f(x), and we select a new set of parameters until the solution satisfies our criterion of "good enough".