PIMS - CSC Seminar: Elena Popovici

Co-optimization is a special type of optimization where the quality of potential solutions (to be optimized) needs to be assessed over a very large number of circumstances (tests).  An important result in co-optimization is that there is “free lunch”, that is, under certain assumptions, some co-optimization algorithms are better than others over all possible problem instances—and we can define optimal algorithms.  The out-of-the-box theoretical formulation of these optimal algorithms appear daunting or completely infeasible from an implementation standpoint.  In this talk I will show how visualization can be used to uncover mathematical properties of optimal algorithms, which can then be formally proven and used to completely or significantly simplify implementation.