Table 2.

Prediction performance benchmarking for the prediction task of “Multi Target Transfer Learning" on the test set.

Property	Data size	Base	MAE of best SC model	MAE of best TL model
KLU	28,056	18.77	11.96	11.37
KAA	28,171	5.234	2.978	2.821
BgOptb	28,163	0.988	0.279	0.251
Deltae	28,155	0.850	0.135	0.120
Encut	28,108	246.25	76.99	83.09
Ehull	27,297	0.131	0.055	0.050
Magoszi	25,844	1.225	0.438	0.405
Magout	25,357	1.176	0.393	0.369
Eps	25,150	3.829	1.462	1.304
PPF	16,250	650.5	543.1	508.6
NPF	16,250	658.1	546.3	493.0
Pem300k	16,763	1.918	1.293	1.111
Nem300k	16,760	1.918	1.282	1.183
PSB	14,439	163.30	68.34	60.53
NSB	14,144	108.69	57.83	53.32
Meps	11,349	4.905	1.926	1.832
MaxM	10,963	285.32	72.66	65.69
MinM	10,930	40.89	24.85	23.51
ETC11	10,839	81.66	37.35	34.03
ETC12	10,759	44.96	19.05	17.15
ETC13	10,846	42.54	15.65	13.90
ETC22	10,832	84.06	36.99	32.13
ETC33	10,856	84.12	38.93	33.89
ETC44	9986	29.55	17.24	14.76
ETC55	9755	26.61	14.90	11.71
ETC66	9739	27.59	15.83	13.81
BulkKV	10,743	49.11	11.83	11.01
ShearGV	10,209	24.28	11.90	11.11
BgMbj	7296	1.911	0.555	0.508
Spillage	3866	0.501	0.379	0.371
SLME	3006	9.439	7.193	6.877
MaxIrM	2302	426.0	108.2	104.6
MinIrM	2268	66.16	49.90	47.14
PMDiEl	2126	5.757	3.221	3.070
PMDi	2126	6.977	3.931	3.761
PMDiIo	2126	2.577	0.847	0.791
PMEij	1123	0.520	0.436	0.415
PMDij	689	46.47	24.43	22.32
Exfoli	557	62.93	59.37	48.11

Only formation energy was used as the source property for this analysis. The table shows the test MAE of the best model selected using Supplementary Table 5 (based on validation MAE) when run on the test set for each of the target materials properties. All the model inputs are based on EF.