How Diverse Initial Samples Help and Hurt Bayesian Optimizers
How Diverse Initial Samples Help and Hurt Bayesian Optimizers
复制标题
多样化的初始样本如何帮助和损害贝叶斯优化器
DOI：
10.1115/1.4063006
复制
发表时间：
2023
影响因子：
3.3
通讯作者：
Fuge, Mark
中科院分区：
文献类型：
--
作者：
Kamrah, Eesh;Ghoreishi, Seyede Fatemeh;Ding, Zijian “Jason”;Chan, Joel;Fuge, Mark
设计研究人员一直在努力进行定量预测，以确定为什么以及何时多样性可能有助于或阻碍设计搜索工作。本文通过研究一种普遍使用的搜索策略-贝叶斯优化（BO）-在具有可修改凸性和难度的2D测试问题上解决了这个问题。具体来说，我们测试如何提供不同的与非多样化的初始样本BO影响其性能在搜索过程中，并介绍了一个快速的排名决定点处理方法计算不同的集，我们需要检测集的高度多样化或非多样化的初始样本。我们最初发现，令我们惊讶的是，多样性似乎不会影响BO，既不会帮助也不会损害优化器的收敛。然而，后续实验揭示了一个关键的权衡。非多样的初始样本加速了底层模型超参数模型构建的后验收敛。相反，不同的初始样本加速了对函数本身的探索--空间探索优势。这两个优势都有助于BO，但方式不同，初始样本多样性直接调节BO如何交易这些优势。事实上，我们表明，固定BO超参数会消除模型构建的优势，导致不同的初始样本总是优于使用非多样样本训练的模型。这些发现揭示了为什么，至少对于BO型优化器，多样性的使用具有混合效果，并警告BO中无处不在的空间填充初始化。在某种程度上，人类使用探索利用搜索策略类似BO，我们的研究结果提供了一个可测试的猜想，为什么以及何时多样性可能会影响人类主题或设计团队的实验。
Design researchers have struggled to produce quantitative predictions for exactly why and when diversity might help or hinder design search efforts. This paper addresses that problem by studying one ubiquitously used search strategy—Bayesian optimization (BO)—on a 2D test problem with modifiable convexity and difficulty. Specifically, we test how providing diverse versus non-diverse initial samples to BO affects its performance during search and introduce a fast ranked-determinantal point process method for computing diverse sets, which we need to detect sets of highly diverse or non-diverse initial samples. We initially found, to our surprise, that diversity did not appear to affect BO, neither helping nor hurting the optimizer’s convergence. However, follow-on experiments illuminated a key trade-off. Non-diverse initial samples hastened posterior convergence for the underlying model hyper-parameters—amodel buildingadvantage. In contrast, diverse initial samples accelerated exploring the function itself—aspace explorationadvantage. Both advantages help BO, but in different ways, and the initial sample diversity directly modulates how BO trades those advantages. Indeed, we show that fixing the BO hyper-parameters removes the model building advantage, causing diverse initial samples to always outperform models trained with non-diverse samples. These findings shed light on why, at least for BO-type optimizers, the use of diversity has mixed effects and cautions against the ubiquitous use of space-filling initializations in BO. To the extent that humans use explore-exploit search strategies similar to BO, our results provide a testable conjecture for why and when diversity may affect human-subject or design team experiments.