Skip to main content

Table 5 Comparison of selective prediction performance in terms of RMSE (%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% 4.799 ± 0.261 \(\mathbf{4.433} \pm \mathbf{0.085}\) \(\mathbf{4.433} \pm \mathbf{0.085}\) \(\mathbf{4.433} \pm \mathbf{0.085}\)
90% 4.129 ± 0.205 4.036 ± 0.130 \(\mathbf{4.003} \pm \mathbf{0.160}\) 4.037 ± 0.161
80% 3.833 ± 0.206 3.796 ± 0.173 3.793 ± 0.182 \(\mathbf{3.765} \pm \mathbf{0.185}\)
70% 3.583 ± 0.249 3.482 ± 0.176 \(\mathbf{3.424} \pm \mathbf{0.196}\) 3.456 ± 0.166
60% 3.382 ± 0.282 3.050 ± 0.261 3.068 ± 0.211 \(\mathbf{3.001} \pm \mathbf{0.184}\)
50% 3.171 ± 0.317 2.653 ± 0.187 2.716 ± 0.168 \(\mathbf{2.605} \pm \mathbf{0.115}\)
40% 2.812 ± 0.218 2.338 ± 0.178 2.503 ± 0.197 \(\mathbf{2.300} \pm \mathbf{0.166}\)
30% 2.518 ± 0.229 2.059 ± 0.245 2.299 ± 0.270 \(\mathbf{2.044} \pm \mathbf{0.235}\)
Suzuki-Miyaura 100% 10.524 ± 0.482 \(\mathbf{9.467} \pm \mathbf{0.459}\) \(\mathbf{9.467} \pm \mathbf{0.459}\) \(\mathbf{9.467} \pm \mathbf{0.459}\)
90% 9.485 ± 0.395 8.632 ± 0.334 8.592 ± 0.338 \(\mathbf{8.540} \pm \mathbf{0.310}\)
80% 8.911 ± 0.373 8.254 ± 0.314 8.146 ± 0.403 \(\mathbf{8.098} \pm \mathbf{0.347}\)
70% 8.473 ± 0.323 7.848 ± 0.329 7.787 ± 0.305 \(\mathbf{7.702} \pm \mathbf{0.397}\)
60% 8.063 ± 0.353 7.260 ± 0.400 7.218 ± 0.343 \(\mathbf{7.160} \pm \mathbf{0.328}\)
50% 7.439 ± 0.470 6.357 ± 0.470 6.503 ± 0.456 \(\mathbf{6.293} \pm \mathbf{0.466}\)
40% 7.236 ± 0.521 5.126 ± 0.306 5.394 ± 0.306 \(\mathbf{4.980} \pm \mathbf{0.250}\)
30% 6.754 ± 0.398 3.968 ± 0.152 4.337 ± 0.257 3.959 ± 0.252