How can you force weights of different layers to be equal in Keras?

Question

Is there a way to force the weights of different layers to be equal during the training of a model in Keras? To be more clear, if I have a model with 5 layers like so:

inputlayer = Dense(units=40, activation='relu', input_dim=20)
hidden1 = Dense(units=40, activation='relu')(inputlayer)
hidden2 = Dense(units=5,activation='relu')(hidden1)
hidden3 = Dense(units=40,activation='relu')(hidden2)
hidden4 = Dense(unites=40,activation='relu')(hidden3)
outputlayer = Dense(units=20,activation='relu')(hidden4)

I would like it so that inputlayer and outputlayer have their weights tied, the same for hidden1 and hidden4, and hidden2 and hidden3. I realize their dimensions are different, ie input layer is(20,40) while outputlayer is (40,20), so I need a way to instantiate the layers so that their weights are tied but transposed as well. How can I do this? Thanks

Answer 1

This is quite easy for the functional API, you just have to do:

layer = Dense(units=40, activation='relu', name="one")
n1 = layer(someInput)
n2 = layer(someOtherInput)

You make one instance of the layer and just give it two different inputs. As you call the layer two times, both instances have the same weights.