notesum.ai

Published at November 27

Transfer Learning for Deep Learning-based Prediction of Lattice Thermal Conductivity

cs.LG

physics.comp-ph

Released Date: November 27, 2024

Authors: L. Klochko¹, M. d'Aquin¹, A. Togo², L. Chaput³

Aff.: ¹Université de Lorraine, LORIA, Nancy F-54000, France; ²Center for Basic Research on Materials (CBRM), National Institute for Materials Science (NIMS), Tsukuba, Ibaraki 305-0047, Japan; ³Université de Lorraine, LEMTA, Nancy F-54000, France; Institut Universitaire de France, 1 rue Descartes, 75231 Paris, France

Arxiv: http://arxiv.org/abs/2411.18259v1

Train on \Test on	Dataset1	Dataset2	MIX	AFLOW
Step 1:
Dataset1	0.82 (0.21)	2.53 (2.17)	1.74 (1.26)	2.16 (1.26)
Dataset2	0.51 (0.09)	0.42 (0.07)	0.43 (0.05)	0.48 (0.05)
MIX	0.69 (0.15)	0.77 (0.15)	0.83 (0.07)	0.97 (0.27)
AFLOW	0.55 (0.34)	1.18 (0.27)	0.93 (0.27)	0.62 (0.33)
Step 2:
Dataset1	0.76 (0.29)	3.09 (2.40)	2.07 (1.40)	2.32 (1.94)
Dataset2	0.55 (0.14)	0.42 (0.06)	0.44 (0.04)	0.52 (0.10)
MIX	0.66 (0.14)	0.75 (0.16)	0.81 (0.11)	1.10 (0.47)
AFLOW	0.44 (0.14)	1.27 (0.48)	0.94 (0.30)	0.50 (0.03)
Step 3:
Dataset1	0.34 (0.15)	1.26 (0.44)	0.87 (0.27)	0.57 (0.10)
Dataset2	0.55 (0.10)	0.78 (0.20)	0.63 (0.18)	0.62 (0.07)
MIX	0.37 (0.14)	0.78 (0.31)	0.69 (0.19)	0.59 (0.06)