Fig. 15
From: Predicting the dissolution kinetics of silicate glasses by topology-informed machine learning
Outcomes of the “topology-informed” machine learning (Model IV-b) using as inputs the number of constraints per atom created by silicon (\({n}_{\mathrm{c}}^{\mathrm{Si}}\)) and aluminum (\({n}_{\mathrm{c}}^{\mathrm{AI}}\)). a Evolution of the relative root square mean square error (RRMSE) of the training and validation sets with respect to the polynomial degree p. The minimum in the RRMSE of the validation set indicates p = 1 as an optimal polynomial degree (i.e., linear model). b Predicted dissolution rate (for p = 1) as a function of the measured dissolution rate. c Coefficients of the polynomial model associated with the \({n}_{\mathrm{c}}^{\mathrm{Si}}\) and \({n}_{\mathrm{c}}^{\mathrm{AI}}\) inputs. Note that, the \({n}_{\mathrm{c}}^{\mathrm{Si}}\) and \({n}_{\mathrm{c}}^{\mathrm{AI}}\) input values are normalized in the training process to ensure that the model coefficients reflect the contribution of each input to the dissolution rate
