Dot-Product Attention

Dot-Product Attention is an attention mechanism where the alignment score function is calculated as:

$$f_{att}\left(\textbf{h}_{i}, \textbf{s}_{j}\right) = h_{i}^{T}s_{j}$$

It is equivalent to multiplicative attention (without a trainable weight matrix, assuming this is instead an identity matrix). Here $\textbf{h}$ refers to the hidden states for the encoder, and $\textbf{s}$ is the hidden states for the decoder. The function above is thus a type of alignment score function.

Within a neural network, once we have the alignment scores, we calculate the final scores/weights using a softmax function of these alignment scores (ensuring it sums to 1).