[1J3-OS-10] OS-10

[1J3-OS-10-04] Quantitative Analysis of Improving Classification Accuracy and Computational Cost for Hyperparameter Optimization

〇Naokatsu Nakazato¹, Masaki Onishi² (1. Graduate School of System and Information Engineering, University of Tsukuba, 2. Artificial Intelligence Research Center, National Institute of Advanced Industrial Science and Technology)

Keywords:Hyperparameter Optimization , Automated Machine Learning(AutoML), ROI(Return on Investment)

In recent years, research called AutoML that aim to automate machine learning is popular in field of deep learning.
Hyperparameter optimization(HPO), a research part of AutoML, strongly impact performance of deep learning algorithms.
However, in order to have exploded computational costs of machine learning, when system optimized hyperparameter introduce to the real world, we need to discuss cost-effectiveness about HPO.
In this paper, we show improvement of classification accuracy and computational cost for HPO of image classification problem.
Also, we estimate economic effects from improvement of classification accuracy and propose quantitative evaluation method of cost-effectiveness of HPO in image classification problem.