A Principle for Global Optimization with Gradients
研究了如何利用梯度信息为多峰目标函数生成非局部二次近似和搜索方向，并在简单算法、CMA-ES和随机重启BFGS中测试了该原理的效果。
Abstract This work demonstrates the utility of gradients for the global optimization of certain differentiable functions with many suboptimal local minima. To this end, a principle for generating non-local quadratic approximants, and the associated search directions, from gradient information of multimodal objective functions is analyzed. Experiments measure the quality of non-local search directions as well as the performance of the principle embedded into a simplistic algorithm, of the covariance matrix adaptation evolution strategy (CMA-ES), and of a randomly reinitialized Broyden-Fletcher-Goldfarb-Shanno (BFGS) method.