view article
Figure 3
The architecture of CPICANN. In each of the one-dimensional convolution layers, n × 1 conv., m and /2 denote the kernel size n, the channel number m and a stride of 2, respectively. In the max-pooling layers, /2 also indicates a stride of 2. Residual connection is indicated by solid lines. The convoluted information is fed into six eight-head self-attention blocks, which scores the input XRD pattern against the 23 073 single-phase patterns.

IUCrJ
Volume 11| Part 4| July 2024| Pages 634-642
ISSN: 2052-2525
https://doi.org/10.1107/S2052252524005323