We study the problem of black-box optimization of a Lipschitz function f defined on a compact subset X of R^d, both via algorithms that certify the accuracy of their recommendations and those that do not. We investigate their sample complexities, i.e., the number of samples needed to either reach or certify a given accuracy epsilon. We start by proving a tighter bound for the well-known DOO algorithm [Perevozchikov, 1990, Munos, 2011] that matches the best existing upper bounds for (more computationally challenging) non-certified algorithms. We then introduce and analyze a new certified version of DOO and prove a matching f-dependent lower bound (up to logarithmic terms) for all certified algorithms. Afterwards, we show that this optimal quantity is proportional to \int_X dx/(max(f) - f(x) + epsilon)^d, solving as a corollary a three-decade-old conjecture by Hansen et al. [1991]. Finally, we show how to control the sample complexity of state-of-the-art non-certified algorithms with an integral reminiscent of the Dudley-entropy integral.
翻译:我们研究Lipschitz 函数f 的黑盒优化问题,Lipschitz 函数在R ⁇ d 的压缩子集X 上定义,通过验证其建议准确性的算法和不精确的算法来研究。我们调查其样本复杂性,即达到或认证给定精度epsilon所需的样本数量。我们首先证明对众所周知的DO 算法[Perevozzchikov,1990年,Munos,2011年]来说,这是与(在计算上更具挑战性的)未经认证的算法相匹配的现有最佳上限。然后我们引入和分析新的经认证的DOO版本,并证明对所有经认证的算法具有匹配的较低约束性(按对数条件)。随后,我们证明这一最佳数量与\int_X dx/(max)- f(x) +epsilon) ⁇ 成正比,作为必然结果解决汉森等人(Hansen 等人) 等人(在30年的测算。[1991年]最后,我们展示如何控制国家非精度综合再定法的样本复杂性。