Run model backwards in Keras

Question

I am currently playing with a Keras framework. And they did some simple classification tests, etc. I would like to find a way to start the network in the reverse order, using the outputs as input and vice versa. Any way to do this?

Answer 1

There is no such thing as “starting the neural network in reverse order”, since the general architecture of the neural network does not determine the processing of data without forwarding . There is, however, a subclass of models that make - generative models that are not part of keras right now. The only thing you can do is create a network that somehow "mimics" the generating process that interests you. But this is a specific method characteristic of the sample, and does not have a general solution.

Answer 2

I'm not sure about this case, but just add a quick solution, you just need to redesign the architecture so that the inputs are then treated as outputs and vice versa.

, ( ):

n_obs, n_feat = 1000, 20
n_labels = 2
n_hidden = 100 # you can make this whatever you want

:

Dense(input_dim = n_feat, output_dim = n_hidden)

:

Dense(input_dim = n_hidden, output_dim = n_labels)

" ", , :

:

Dense(input_dim = n_labels, output_dim = n_hidden)

:

Dense(input_dim = n_hidden, output_dim = n_feat)

, , . , . ?

, , (.. ), , .

, , Neural Networks, .