Skip to main content
. 2018 Mar 8;3(3):2825–2836. doi: 10.1021/acsomega.7b02045

Figure 11.

Figure 11

Skip-gram neural network architecture for generating fragment vectors. The target fragment is shown in red, and its one-hot vector is fed as the input. The objective is to get the highest probability for the context fragment (shown in blue) in the softmax output layer.