3次元畳み込みニューラルネットワークの転移学習を用いたブロック共重合体の応力ひずみ曲線予測

抄録

The simulation to obtain stress-strain curves of block copolymers requires large computational resources. As an alternative to simulation, a method for high-throughput prediction of stress-strain curves using a three-dimensional convolutional neural network has been reported. In this study, we incorporated shortcut coupling into the neural network and performed pre-training and transfer learning in a step-by-step manner to successfully predict the stress-strain curve with high accuracy while reducing the training cost.