Attention in Keras : How to add different attention mechanism in keras Dense layer?

Question

I am new in Keras and I am trying to build a simple autoencoder in keras with attention layers :

Here what I tried :

data = Input(shape=(w,), dtype=np.float32, name='input_da')
noisy_data = Dropout(rate=0.2, name='drop1')(data)

encoded = Dense(256, activation='relu',
            name='encoded1', **kwargs)(noisy_data)
encoded = Lambda(mvn, name='mvn1')(encoded)

encoded = Dense(128, activation='relu',
            name='encoded2', **kwargs)(encoded)

encoded = Lambda(mvn, name='mvn2')(encoded)
encoded = Dropout(rate=0.5, name='drop2')(encoded)


encoder = Model([data], encoded)
encoded1 = encoder.get_layer('encoded1')
encoded2 = encoder.get_layer('encoded2')


decoded = DenseTied(256, tie_to=encoded2, transpose=True,
            activation='relu', name='decoded2')(encoded)
decoded = Lambda(mvn, name='new_mv')(decoded)


decoded = DenseTied(w, tie_to=encoded1, transpose=True,
            activation='linear', name='decoded1')(decoded)

And it looks like this:

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
data (InputLayer)            (None, 2693)              0         
_________________________________________________________________
drop1 (Dropout)              (None, 2693)              0         
_________________________________________________________________
encoded1 (Dense)             (None, 256)               689664    
_________________________________________________________________
mvn1 (Lambda)                (None, 256)               0         
_________________________________________________________________
encoded2 (Dense)             (None, 128)               32896     
_________________________________________________________________
mvn2 (Lambda)                (None, 128)               0         
_________________________________________________________________
drop2 (Dropout)              (None, 128)               0         
_________________________________________________________________
decoded2 (DenseTied)         (None, 256)               256       
_________________________________________________________________
mvn3 (Lambda)                (None, 256)               0         
_________________________________________________________________
decoded1 (DenseTied)         (None, 2693)              2693      
=================================================================

Where I can add attention layer in this model? should I add after first encoded_output and before second encoded input?

encoded = Lambda(mvn, name='mvn1')(encoded)

    Here?

encoded = Dense(128, activation='relu',
            name='encoded2', **kwargs)(encoded)

also I was going though this beautiful lib :

https://github.com/CyberZHG/keras-self-attention

They have implemented various types of attention mechanisms but it's for sequential models. How I can add those attention in my model?

I tried with very simple attention :

encoded = Dense(256, activation='relu',
        name='encoded1', **kwargs)(noisy_data)


encoded = Lambda(mvn, name='mvn1')(encoded)

attention_probs = Dense(256, activation='softmax', name='attention_vec')(encoded)
attention_mul = multiply([encoded, attention_probs], name='attention_mul')
attention_mul = Dense(256)(attention_mul)

print(attention_mul.shape)

encoded = Dense(128, activation='relu',
        name='encoded2', **kwargs)(attention_mul)

is it at right place and can I add any other attention mechanism with this model?

Answer 1

I guess what you're doing is a correct way of adding attention, because attention in itself is nothing but can be visualized as weights of a dense layer. Also, I guess applying attention just after encoder is the right thing to do, as it will apply attention to the most "informative" part of the data distribution necessary for your task.