Study on Acquisition Functions

Konferenz: ICMLCA 2021 - 2nd International Conference on Machine Learning and Computer Application
17.12.2021 - 19.12.2021 in Shenyang, China

Tagungsband: ICMLCA 2021

Seiten: 5Sprache: EnglischTyp: PDF

Persönliche VDE-Mitglieder erhalten auf diesen Artikel 10% Rabatt

Autoren:
Sun, Yiming (Wellington School Ayr, Ayrshire, UK)
Wu, Zhengyang (Guanghua Cambridge International School, Shanghai, China)
Gu, Haoyu (Shanghai Experimental Foreign Language School, Shanghai, China)
Zhou, Yuguang (International Department, The Affiliated High School of SCNU, Guangzhou, China)

Inhalt:
This paper is about statistical Machine Learning’s hyperparameter tuning, using Bayesian Optimization and its Acquisition Functions (AF). We coded on Python and carried out a simple SVM classification task. In conclusion, by comparison the most balanced in exploration and exploitation and well performed AF is the Expected Improvement (EI) AF. Then, an enhanced Acquisition function is proposed, the Modified EI, including noise. Future attempts are needed to be carried out on different task types to prove its superiority.