Skip to main content

Table 4 Comparison of selective prediction performance in terms of MAE (%p)

From: Uncertainty-aware prediction of chemical reaction yields with graph neural networks

Dataset Coverage YieldBERT-DA Proposed (\(\lambda = 0.1\))
Aleatoric Epistemic Total Pred. Var.
Buchwald-Hartwig 100% 3.090 ± 0.118 \(\mathbf{2.920} \pm \mathbf{0.056}\) \(\mathbf{2.920} \pm \mathbf{0.056}\) \(\mathbf{2.920} \pm \mathbf{0.056}\)
90% 2.733 ± 0.099 2.684 ± 0.050 \(\mathbf{2.669} \pm \mathbf{0.056}\) 2.683 ± 0.061
80% 2.534 ± 0.082 2.518 ± 0.064 2.514 ± 0.063 \(\mathbf{2.505} \pm \mathbf{0.065}\)
70% 2.357 ± 0.092 2.302 ± 0.067 \(\mathbf{2.292} \pm \mathbf{0.067}\) 2.293 ± 0.064
60% 2.191 ± 0.103 2.056 ± 0.099 2.070 ± 0.064 \(\mathbf{2.041} \pm \mathbf{0.069}\)
50% 2.020 ± 0.105 1.820 ± 0.093 1.847 ± 0.075 \(\mathbf{1.803} \pm \mathbf{0.061}\)
40% 1.824 ± 0.106 1.593 ± 0.086 1.672 ± 0.081 \(\mathbf{1.582} \pm \mathbf{0.077}\)
30% 1.560 ± 0.098 \(\mathbf{1.368} \pm \mathbf{0.112}\) 1.509 ± 0.115 1.372 ± 0.111
Suzuki-Miyaura 100% 6.598 ± 0.270 \(\mathbf{6.116} \pm \mathbf{0.223}\) \(\mathbf{6.116} \pm \mathbf{0.223}\) \(\mathbf{6.116} \pm \mathbf{0.223}\)
90% 5.902 ± 0.247 5.589 ± 0.178 5.575 ± 0.191 \(\mathbf{5.542} \pm \mathbf{0.178}\)
80% 5.415 ± 0.242 5.298 ± 0.174 5.269 ± 0.210 \(\mathbf{5.219} \pm \mathbf{0.192}\)
70% 5.052 ± 0.211 5.018 ± 0.196 4.966 ± 0.183 \(\mathbf{4.939} \pm \mathbf{0.208}\)
60% 4.690 ± 0.181 4.641 ± 0.218 4.579 ± 0.140 \(\mathbf{4.570} \pm \mathbf{0.188}\)
50% 4.213 ± 0.214 4.025 ± 0.252 4.064 ± 0.179 \(\mathbf{3.989} \pm \mathbf{0.203}\)
40% 3.921 ± 0.188 3.245 ± 0.140 3.372 ± 0.111 \(\mathbf{3.195} \pm \mathbf{0.145}\)
30% 3.549 ± 0.120 \(\mathbf{2.510} \pm \mathbf{0.093}\) 2.701 ± 0.118 2.514 ± 0.115