Amal FLIMINE
Engineering student at ENAC, Toulouse, France. Specialized in Air Operations & Safety with a double Master degree in Operational Research.
Interested in different fields of Optimization, I joined the Alliance team in April 2022 at the GERAD laboratory under the supervision of Sébastien Le Digabel, Charles Audet and Christophes Tribes.
Contact : amal.flimine@polymtl.ca
Search strategies in MADS: Line search & DiRect based methods
MADS is a derivative-free optimization algorithm that includes an exploration of the design space. It succeeds to direct-search algorithms such as Coordinate Search and Generalized Pattern Search, allowing for an infinite set of directions, and dealing with general inequality constraints. 1) MADS can use components to break free from local minimisers during the Search step and accelerate convergence to a better minimizer.
My project is to implement new search strategies in MADS based on:
- Line-search, which simply takes a step from the current solution xk of length tk relative to a direction dk. The next incumbent solution needs to satisfy a minimal decrease on the criteria. An exploration of dfo line-search algorithms is necessary to implement this subroutine in Speculative Search.
- DiRect, a shorthand for Dividing Rectangles, which is a hypercube-partitioning algorithm that aims to locate a feasible sub-region containing the global minimizer. 2)