Sample data from Deep Neural Network

User 3843 | 4/13/2016, 8:57:29 AM

Hi,

I have a network trained with some data. I would like to sample data from initial input data.

Mainly, calculate probability vector P given an input data (partial data). Mainly, to generate data from the trained Deep Neural Network.

Thanks, regards

Comments

User 4 | 4/13/2016, 6:33:12 PM

Hi @brookm297, I'm not sure what you mean. Can you elaborate on what you are trying to do and the problem you are encountering?


User 3843 | 4/13/2016, 6:56:05 PM

Hi,

I have a recurrent deep neural network trained on some image data.

For an initial input, I would like to get the output probability through the trained network given an arbitray input (and then use later this probability to sample output data).

Is it possible in Graphlab ?

Thanks


User 4 | 4/13/2016, 8:29:22 PM

Hi @brookm297, it sounds like you are asking about a generative neural network model. Unfortunately I don't think GraphLab Create currently supports this use of neural network models, but if this is a feature you would like, please submit a post in the Feature Requests category so that others may comment on it there as well. Thanks!