シミュレーションと機械学習の組み合わせによる高速近似解析手法構築の能動学習を用いた効率化に関する研究

抄録

In recent years, AI has been adopted in the manufacturing industry, and machine learning-based approximator systems are expected to be used for shortening the development time of products. Although it is generally accepted that big data is essential for training data, it takes a long time to collect the data obtained from simulations, so that a high-performance approximator must be built with a minimum amount of training data.Although active learning has been studied in various fields to reduce the data collection cost, most of the research is on the classification problem, and its adaptation to the regression problem is not advanced. In this study, we construct a simulation approximator for this regression problem with a minimal amount of training data, using expressive neural networks to approximate unknown functions.