From Conventional Machine Learning to AutoML
Ziqiao Weng · Journal of Physics Conference Series · 2019
Machine Learning has enabled conspicuous progress over the past decade on various areas, such as image analysis, computer vision, natural language processing, quantitative finance, etc. This paper presents an overview of Machine Learning algorithms from conventional Machine Learning to Deep Learning then to Automated Machine Learning (AutoML). In the first two sections, a few classical and distinct algorithms will be discussed in the same structure: 1) the principle of algorithm, 2) specific applications or improvements, and 3) contributions or limitations. At the last section, the state-of-the-art AutoML algorithms will be briefly introduced along with their innovation, which achieve unexpected higher efficiency. It turns out that there is a growing interest in AutoML methods. In the future, more and more novel neural architectures will be proposed and applications to real-world scenario will be expanded.