journal menu
![[Figure 2]](vb5091fig2mag.jpg)
|
|
Figure 2
CNNs for (a) classifying only the space group (SpgNet) and (b) predicting the space group and cell volume using a multi-task learning approach (SpgVolNet) from XRD profiles. The diffraction intensity Ihkl as a function of the reciprocal interplanar distance 1/d divided into 10001 bins is loaded to the input layer, depicted in grey. The convolutional layers are shown in blue, with f channels of kernel size k and stride s. The activation functions are shown in orange. The dropout layers (in green) are inserted to control overfitting in (a). The average pooling layers are shown in red with kernel size k and stride s. The layer in white converts its input to one dimension without changing the values. The layers in yellow are fully connected. In the output layers in both (a) and (b), the identity function is adopted as the activation function. |
![[Figure 2]](vb5091fig2.jpg)
 | JOURNAL OF APPLIED CRYSTALLOGRAPHY |
ISSN: 1600-5767
access
