Table 1 Predicted probabilities in % on the experimental data from several models trained with different simulated training data sets.
From: A convolutional neural network for defect classification in Bragg coherent X-ray diffraction
Experimental example | P10 mixed 1 | SixS Screw 1 | P10 no defect 1 | P10 no defect 2 | SixS Screw 2 | P10 mixed 2 | SixS no defect | P10 mixed 3 | P10 no defect 3 |
|---|---|---|---|---|---|---|---|---|---|
Training data set | |||||||||
Pt-unrelaxed small crystals CD | p 99 s 1 e 0 | p 0 s 3 e 97 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 0 s 1 e 99 | p 97 s 2 e 1 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 100 s 0 e 0 |
Pt-relaxed small crystals CD | p 99 s 1 e 0 | p 0 s 100 e 0 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 0 s 69 e 31 | p 72 s 18 e 10 | p 72 s 16 e 12 | p 92 s 6 e 2 | p 100 s 0 e 0 |
Pt-relaxed big crystals CD | p 99 s 0 e 1 | p 0 s 100 e 0 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 25 s 38 e 37 | p 99 s 0 e 1 | p 99 s 0 e 1 | p 99 s 0 e 1 | p 100 s 0 e 0 |
Multi-elements relaxed big crystals CD | p 98 s 0 e 2 | p 0 s 100 e 0 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 0 s 98 e 2 | p 88 s 5 e 8 | p 99 s 0 e 1 | p 94 s 2 e 3 | p 100 s 0 e 0 |
Multi-elements relaxed big crystals RPD | p 9 s 53 e 38 | p 0 s 36 e 64 | p 100 s 0 e 0 | p 100 s 0 e 0 | p 0 s 8 e92 | p 0 s 58 e 42 | p 99 s 1 e 0 | p 0 s 63 e 37 | p 100 s 0 e 0 |
Multi-elements-relaxed big crystals 75% CD, 25% RPD | p 0 s 94 e 6 | p 0 s 100 e 0 | p 98 s 2 e 0 | p 99 s 1 e 0 | p 0 s 97 e 3 | p 0 s 84 e 16 | p 75 s 16 e 9 | p 0 s 82 e 18 | p 100 s 0 e 0 |
