Machine Learning Curriculums Generated by Classifier Ensembles
Machine Learning Curriculums Generated by Classifier Ensembles
复制标题
DOI：
10.1109/cspa57446.2023.10087822
复制
发表时间：
2023-03
期刊：
影响因子：
--
通讯作者：
Tzu-Jui Huang;K. Kameyama
中科院分区：
文献类型：
--
作者：
Tzu-Jui Huang;K. Kameyama
图像分类是一项基本任务，试图将图像分类为它们所属的类别。 CNN 由于其灵活性和稳定性而被广泛用于分类。本文将课程学习视为一种辅助方法，决定 CNN 的训练过程。课程学习通常被定义为一种将训练数据集划分并排序为几个不同学习难度级别的方法。本文利用集成分类的思想，基于不同机器学习模型的共识，按照顺序进行训练。因此，与使用狗脸识别和 CIFAR-10 数据集的单阶段 CNN 学习相比，所提出的集成生成课程学习模型具有更高的准确性。
Image classification is a fundamental task that attempts to classify the images into the classes they belong to. CNN is commonly used for the classification due to its flexibility and stability. This paper treats Curriculum Learning as an assistant method that decides the procedure the CNN ought to be trained. Curriculum Learning is generally defined as a method that divides and orders the training datasets into several different levels of learning difficulties. This paper utilizes the idea of ensemble classification that trains according to the order based on the consensus of different machine learning models. As a result, the proposed Ensemble-Generated Curriculum Learning model achieves higher accuracy compared to that of a single-stage CNN learning with dog face recognition and CIFAR-10 datasets.