03:30 PM - 03:55 PM
Use of Models with the MADS Algorithm for Blackbox Optimization
Blackbox optimization occurs when the functions representing the objective and constraints have no exploitable structure, including available derivatives. Such functions are typically evaluated via computer simulations. The Mesh Adaptive Direct Search algorithm (MADS) is a directional direct search method specifically designed for such problems. This work exploits the algorithm flexibility by incorporating models of the objective and of the constraints in order to guide the search. In particular, local quadratic models and global treed Gaussian processes are considered and tested.
03:55 PM - 04:20 PM
Optimization Under Unknown Constraints
We consider optimization for computer experiments under unknown constraints, i.e., when simulation is required to determine real-valued responses and check constraints. We develop surrogates to approximate both simulator outputs. A new integrated improvement criterion recognizes that responses that violate the constraint are still informative about the function, and thus potentially useful in the optimization. We illustrate our approach on a problem from health care policy.
04:20 PM - 04:45 PM
Optimization of the Location and Frequency of Emitters in a Wireless Communications Network
When supplying a territory with wireless communications services, such as mobile phones, you need to place emitting antennas on the terrain. In order to cover it entirely and serve a maximum of users, the antennas locations have to be optimized. Also, since antennas emitting on the same frequency interfere in a strong way, the frequency of each antenna has to be chosen carefully. These two problems are here treated together with black box optimization using the MADS algorithm and metaheuristics.
04:45 PM - 05:10 PM
Snow Water Equivalent Estimation Using Blackbox Optimization
Accurate measurements of snow water equivalent is an important factor in hydroelectric power generation. SWE is estimated via kriging on measures obtained by snow monitoring devices. The question of positioning the devices in order to minimize the kriging interpolation error is formulated as a blackbox optimization problem. The 2D nature of the variables is exploited to form groups.