TY - GEN
T1 - Edge Based Architecture for Total Energy Regression Models for Computational Materials Science
AU - Yeo, Kangmo
AU - Jeong, Sukmin
AU - Kim, Soo Hyung
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.
PY - 2023
Y1 - 2023
N2 - It is widely acknowledged that artificial intelligence (AI) technology has been extensively applied and has achieved remarkable advancements in various fields. The field of computational materials science has also embraced AI techniques in diverse ways. Today, computational materials science plays a crucial role in the development of cutting-edge materials, including pharmaceuticals, catalysts, semiconductors, and batteries. One significant task in this field is the regression of the total energy of atomic structures that form various materials. In this study, we propose a modified model architecture aimed at improving the performance of existing total energy regression models. Traditional total energy regression models calculate the total energy by summing the energies of individual nodes represented in the atomic structure graph. However, our approach suggests a modified architecture that not only predicts the energy for nodes but also incorporates energy prediction for edges in the graph. This novel architecture achieved a 3.9% reduction in energy error compared to the base model. Moreover, its simplicity provides the advantage of general applicability to other total energy regression models.
AB - It is widely acknowledged that artificial intelligence (AI) technology has been extensively applied and has achieved remarkable advancements in various fields. The field of computational materials science has also embraced AI techniques in diverse ways. Today, computational materials science plays a crucial role in the development of cutting-edge materials, including pharmaceuticals, catalysts, semiconductors, and batteries. One significant task in this field is the regression of the total energy of atomic structures that form various materials. In this study, we propose a modified model architecture aimed at improving the performance of existing total energy regression models. Traditional total energy regression models calculate the total energy by summing the energies of individual nodes represented in the atomic structure graph. However, our approach suggests a modified architecture that not only predicts the energy for nodes but also incorporates energy prediction for edges in the graph. This novel architecture achieved a 3.9% reduction in energy error compared to the base model. Moreover, its simplicity provides the advantage of general applicability to other total energy regression models.
KW - Atomic structure
KW - Computational materials science
UR - https://www.scopus.com/pages/publications/85177450542
U2 - 10.1007/978-3-031-47665-5_9
DO - 10.1007/978-3-031-47665-5_9
M3 - Conference paper
AN - SCOPUS:85177450542
SN - 9783031476648
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 106
EP - 112
BT - Pattern Recognition - 7th Asian Conference, ACPR 2023, Proceedings
A2 - Lu, Huimin
A2 - Blumenstein, Michael
A2 - Cho, Sung-Bae
A2 - Liu, Cheng-Lin
A2 - Yagi, Yasushi
A2 - Kamiya, Tohru
PB - Springer Science and Business Media Deutschland GmbH
T2 - 7th Asian Conference on Pattern Recognition, ACPR 2023
Y2 - 5 November 2023 through 8 November 2023
ER -