変分オートエンコーダを用いた形状のデータマイニングと形状創出

書誌事項

タイトル別名
  • Making Research Paper

抄録

<p>We propose variational autoencoder (VAE) based approach for generating shapes that satisfy requirements. In mechanical design process, designers at first design rough drawings that satisfy requirements, and then the designer start modification or optimization of the shape. It is not easy for non-skilled designers to draw the rough drawings in short time, and even sometimes it is not easy for skilled designers. In the present study, we use conditional variational autoencoders (CVAE) to generate shapes. Also, we compare results by CVAE models using normal distribution and von Mises-Fisher distribution. We apply CVAE model to 2-dimensional laminar airfoil dataset and compare results.</p>

収録刊行物

詳細情報 詳細情報について

問題の指摘

ページトップへ