Presentation Information
[20a-P04-4]Active learning in neural networks training for catalyst energy prediction
〇Yasufumi Sakai1, Thang Dang1, Koichi Shirahata1, Atsushi Ishikawa2 (1.Fujitsu Ltd., 2.Tokyo Tech.)
Keywords:
neural network,active learning,energy prediction for materials
Recently, many earlier works have been proposed to utilize neural networks (NN) for predicting the energy of material for material search instead of calculating the energy by DFT with high computational cost. On the other hand, NN training generally requires a large amount of training data, i.e., a large amount of DFT calculations. In this work, we evaluate a NN training method utilizing active learning for improving the NN accuracy in the situations where the number of training data is small. The active learning expects automatically selects effective data for improving the accuracy of NNs.