Attention Mechanisms

Additive Attention

Introduced by Bahdanau et al. in Neural Machine Translation by Jointly Learning to Align and Translate

Additive Attention, also known as Bahdanau Attention, uses a one-hidden layer feed-forward network to calculate the attention alignment score:

$$f_{att}\left(\textbf{h}_{i}, \textbf{s}_{j}\right) = v_{a}^{T}\tanh\left(\textbf{W}_{a}\left[\textbf{h}_{i};\textbf{s}_{j}\right]\right)$$

where $\textbf{v}_{a}$ and $\textbf{W}_{a}$ are learned attention parameters. 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. We can use a matrix of alignment scores to show the correlation between source and target words, as the Figure to the right shows.

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

Source: Neural Machine Translation by Jointly Learning to Align and Translate

Papers


Paper Code Results Date Stars

Tasks


Task Papers Share
Speech Synthesis 42 11.48%
Sentence 28 7.65%
Reinforcement Learning (RL) 15 4.10%
Text-To-Speech Synthesis 15 4.10%
Combinatorial Optimization 13 3.55%
Language Modelling 10 2.73%
Speech Recognition 8 2.19%
Starcraft II 7 1.91%
Translation 7 1.91%

Categories