A Feature-Free Approach to Automated Algorithm Selection
A Feature-Free Approach to Automated Algorithm Selection
复制标题
一种无特征的自动算法选择方法
DOI：
10.1145/3583133.3595832
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Alissa M
中科院分区：
文献类型：
--
作者：
Alissa M
本文总结了2023年1月在启发式杂志上发表的无特征算法选择领域的最新工作，标题为“自动算法选择：从基于特征的方法到无特征的方法”。具体地说，我们考虑在数据中封装有隐式顺序信息的领域，例如在线装箱。我们训练了两种类型的递归神经网络（RNN）来预测在线装箱中的包装启发式，该启发式使用物品大小序列作为输入，即没有衍生出特征来描述实例。这与算法选择的典型方法形成了对比。RNN方法被证明能够在80.88到97.63%的实例上实现oracle性能的5%以内，具体取决于数据集。它们也被证明优于使用衍生特征训练的经典机器学习模型。为了解释观察到的结果，我们建议当实例表现出一些隐式结构时，我们的方法表现良好。为了证明这一点，我们生成了14个具有可控结构水平的新数据集，这表明在算法选择产生效益之前，需要一个关键的结构阈值。
This article summarises recent work in the domain of feature-free algorithm selection that was published in the Journal of Heuristics in January 2023, with the title 'Automated Algorithm Selection: from Feature-Based to Feature-Free Approaches'. Specifically, we consider domains in which there is implicit sequential information encapsulated in the data, e.g., in online bin-packing.We train two types of recurrent neural networks (RNN) to predict a packing heuristic in online bin-packing that use the sequence of item-sizes as input, i.e. no features are derived to describe the instance. This contrasts to typical approaches to algorithm-selection. The RNN approaches are shown to be capable of achieving within 5% of the oracle performance on between 80.88 and 97.63% of the instances, depending on the dataset. They are also shown to outperform classical machine learning models trained using derived features. In order to explain the observed result, we suggest that our methods perform well when the instances exhibit some implicit structure. To provide evidence for this, 14 new datasets with controllable levels of structure are generated, indicating that a critical threshold of structure is required before algorithm-selection delivers benefit.
使用深度学习进行算法选择，无需特征提取
DOI：
--
发表时间：
2019
期刊：
Annual Conference on Genetic and Evolutionary Computation
影响因子：
--
作者：
M. Alissa;Kevin Sim;E. Hart
通讯作者：
E. Hart