Fig. 5: Transferability of the ML method across different sample sizes.
From: Predicting plastic deformation of crystalline materials by deciphering acoustic emission
The R2 score for different combinations of 4-experiment training sets and 1-experiment test sets. The R2 scores are evaluated for the fine-scale predictions. The combinations with the test set of the same micropillar diameter (8 μm, 16 μm or 32 μm) are arranged in distinct three panels. The colour of the violin plots in each panel indicates the test sample size (see colourbar). The size composition of the training sets are showed by the miniature pie charts which use the same colour-coding. The striped slices denote a training experiment that is either of one diameter or of another. In each scenario two point clouds and violin plots indicate the performance for the frequency independent and frequency dependent approaches. Each data point corresponds to the performance score R2 of an individual training-test set pair. The mean R2 scores are indicated by the orange and purple markers in the two cases of frequency independent and frequency dependent approaches. Generally, it can be observed that, even though, the prediction score improves as the similarity of the training and the testing data increases, the model performs reasonably well even if experiments of the same pillar size as in the test experiment were not shown. This observation holds regardless of the inclusion/exclusion of frequency information.
