CODEWITHSUNDEEP
rh2o
RISHAV SINGH
RISHAV SINGH

Reputation: 21

H2O.deeplearning neural network

Few Doubts related to h2o.deeplearning:

  1. Is it a deep neural network or deep belief neural network(which uses Restricted Boltzman Machine - RBM)
  2. No where in the documentation of H2o,RBM is discussed.They are using SGD Stochastic gadient descent (which is being used by normal neural network).
  3. If we are not using autoencoders and RBM then can we call it a deep belief neural network.

H2o is really very interesting, but with these doubts it will be very difficult to use it.

Upvotes: 2

Views: 812

Answers (1)

ikkjo
ikkjo

Reputation: 785

I am not an expert in the area, but I do not think h2o is claiming to implement a deep belief network (DBN). I think they implement a deep neural network (DNN) with feedforward. The documentation (https://github.com/h2oai/h2o-3/blob/master/h2o-docs/src/booklets/v2_2015/PDFs/online/DeepLearning_Vignette.pdf) also states as much; in the introduction it says topic include:

Building deep neural nets in H2O

This will explain the lack of RBMs you describe.

For some more detail on the distinction between DNNs and DBNs, also see the accepted answer here: https://stats.stackexchange.com/questions/51273/what-is-the-difference-between-a-neural-network-and-a-deep-belief-network/59854#59854

Regarding autoencoders, they can be used though. For details, see section 7 of the pdf linked above.

Upvotes: 2

Related Questions