Combining F-Race and mesh adaptive direct search for automatic algorithm configuration
暂无分享,去创建一个
In this article, we tackle the automatic algorithm configuration problem (ACP), finding the best configuration of an algorithm such that some measure of its performance is optimized. We study the mesh adaptive direct search (MADS) method for the ACP. MADS is an iterative algorithm for global optimization. It does not require any infomation of the evaluation function, therefore the ACP can be regarded as a black-box for evaluation. To handle the stochastic nature of the ACP, we adopted F-Race, to adaptively allocate the evaluation budgets among a population of candidate configurations. We compare the hybrid of MADS and F-Race (MADS/F-Race) to MADS with certain fixed numbers of evaluations, and demonstrate the advantage and robustness of MADS/F-Race over its counterparts in solving the ACP.
[1] Mauro Birattari,et al. Tuning Metaheuristics - A Machine Learning Perspective , 2009, Studies in Computational Intelligence.
[2] CHARLES AUDET,et al. Finding Optimal Algorithmic Parameters Using Derivative-Free Optimization , 2006, SIAM J. Optim..
[3] Thomas Stützle,et al. A Racing Algorithm for Configuring Metaheuristics , 2002, GECCO.
[4] Charles Audet,et al. Mesh Adaptive Direct Search Algorithms for Constrained Optimization , 2006, SIAM J. Optim..