Skip to main content
. 2023 Oct 9;8(42):39759–39769. doi: 10.1021/acsomega.3c05753

Figure 2.

Figure 2

Comparison between the standard self-attention (left) and GeoAttention (right). The key differences of two architectures are that the GeoAttention 1) used the RBFemb to represent distances of atom pairs and 2) removed the softmax function existing in the standard self-attention.