Gated attention neural networks
A system including an attention neural network that is configured to receive an input sequence and to process the input sequence to generate an output is described. The attention neural network includes: an attention block configured to receive a query input, a key input, and a value input that are derived from an attention block input. The attention block includes an attention neural network layer configured to: receive an attention layer input derived from the query input, the key input, and the value input, and apply an attention mechanism to the query input, the key input, and the value input to generate an attention layer output for the attention neural network layer; and a gating neural network layer configured to apply a gating mechanism to the attention block input and the attention layer output of the attention neural network layer to generate a gated attention output.
1. A system comprising one or more computers and one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to implement an attention neural network that is configured to receive a network input and to process the network input to generate an output, the attention neural network comprising:
an attention block configured to receive a query input, a key input, and a value input that are derived from an attention block input, the attention block comprising:
an attention neural network layer configured to:
receive an attention layer input derived from the query input, the key input, and the value input, and
apply an attention mechanism to the attention layer input to generate an attention layer output for the attention neural network layer; and
a gating neural network layer configured to apply a gating mechanism to the attention block input and the attention layer output of the attention neural network layer to generate a gated attention output.
2. The system of claim 1 , wherein the attention block further comprises a first layer normalization layer configured to apply a layer normalization operation to the query input, the key input, and the value input to generate a normalized query input, a normalized key input, and a normalized value input, and wherein the attention layer input comprises the normalized query input, the normalized key input, and the normalized value input.
3. The system of claim 1 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid modulation to the attention block input to generate a first sigmoid modulated output; and
combining the first sigmoid modulated output with the attention layer output to generate the gated attention output.
4. The system of claim 1 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid modulation to the attention layer output to generate a second sigmoid modulated output, and
combining the second sigmoid modulated output with the attention block input to generate the gated attention output.
5. The system of claim 1 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises: computing a convex combination of the attention block input and the attention layer output using a sigmoid weighting to generate the gated attention output.
6. The system of claim 1 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid and a tan h activation on the attention layer output to generate a sigmoid-tan h output, and
combining the sigmoid-tan h output with the attention block input to generate the gated attention output.
7. The system of claim 1 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a gated recurrent unit on the attention block input and the attention layer output.
8. The system of claim 1 , wherein the attention block further comprise:
a second layer normalization layer configured to apply a second layer normalization operation to the gated attention output to generate a normalized-gated attention output,
one or more feedforward neural network layers configured to apply one or more transformations to the normalized-gated attention output to generate a temporary attention block output, and
a second gating neural network layer configured to apply a second gating mechanism to the temporary attention block output and the gated attention output to generate a final attention block output for the attention block.
9. The system of claim 1 , wherein the attention mechanism is a self-attention mechanism.
10. The system of claim 1 , wherein the attention mechanism is a masked self-attention mechanism.
11. One or more non-transitory computer storage media storing instructions that, when executed by one or more computers, cause the one or more computers to perform operations for processing an attention block input of an attention block of an attention neural network, the operations comprising:
receiving a query input, a key input, and a value input that are derived from the attention block input;
receiving, by an attention neural network layer of an attention block, an attention layer input derived from the query input, the key input, and the value input;
applying, using the attention neural network layer, an attention mechanism to the attention layer input to generate an attention layer output for the attention neural network layer; and
applying, using a gating neural network layer of the attention block, a gating mechanism to the attention block input and the attention layer output of the attention neural network layer to generate a gated attention output.
12. The one or more non-transitory computer storage media of claim 11 , wherein the operations further comprise:
applying a layer normalization operation to the query input, the key input, and the value input to generate a normalized query input, a normalized key input, and a normalized value input, and wherein the attention layer input comprises the normalized query input, the normalized key input, and the normalized value input.
13. The one or more non-transitory computer storage media of claim 11 , wherein the operations further comprise:
applying a second layer normalization operation to the gated attention output to generate a normalized-gated attention output;
applying one or more transformations to the normalized-gated attention output to generate a temporary attention block output; and
applying, using a second gating neural network layer of the attention block, a second gating mechanism to the temporary attention block output and the gated attention output to generate an attention block output for the attention block.
14. A computer-implemented method for processing an attention block input of an attention block of an attention neural network, the method comprising:
receiving a query input, a key input, and a value input that are derived from the attention block input;
receiving, by an attention neural network layer of an attention block, an attention layer input derived from the query input, the key input, and the value input;
applying, using the attention neural network layer, an attention mechanism to the attention layer input to generate an attention layer output for the attention neural network layer; and
applying, using a gating neural network layer of the attention block, a gating mechanism to the attention block input and the attention layer output of the attention neural network layer to generate a gated attention output.
15. The method of claim 14 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid modulation to the attention block input to generate a first sigmoid modulated output; and
combining the first sigmoid modulated output with the attention layer output to generate the gated attention output.
16. The method of claim 14 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid modulation to the attention layer output to generate a second sigmoid modulated output, and combining the second sigmoid modulated output with the attention block input to generate the gated attention output.
17. The method of claim 14 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises: computing a convex combination of the attention block input and the attention layer output using a sigmoid weighting to generate the gated attention output.
18. The method of claim 14 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a sigmoid and a tan h activation on the attention layer output to generate a sigmoid-tan h output, and
combining the sigmoid-tan h output with the attention block input to generate the gated attention output.
19. The method of claim 14 , wherein applying the gating mechanism on the attention block input and the attention layer output comprises:
applying a gated recurrent unit on the attention block input and the attention layer output.
20. The method of claim 14 , further comprising:
applying a layer normalization operation to the gated attention output to generate a normalized-gated attention output;
applying one or more transformations to the normalized-gated attention output to generate a temporary attention block output; and
applying, using a second gating neural network layer of the attention block, a second gating mechanism to the temporary attention block output and the gated attention output to generate an attention block output for the attention block.