What's the difference between deep belief network and deep convex network?


Deep convex network is relatively new architecture of deep neural networks, which has been developed in order to overcome the scalability limitations of deep belief networks (DBN). You can read more on technical details of the architecture and its performance in research papers, for example, in this paper as well as a relevant later paper, both from Microsoft Research.

It might be beneficial to read more about DBN on this fascinating peer-reviewed open-access site, as it's more comprehensive than Wikipedia on the topic. In my opinion, even more comprehensive and interesting overview of deep learning architectures in AI can be found in this technical report.

  • $\begingroup$ interesting.....i'll read those papers.....is there any open source implementation for deep convex net? $\endgroup$ – hadooper Dec 17 '14 at 13:48
  • $\begingroup$ @hadooper: Thanks for upvoting, glad you find it helpful. Don't know about the open source implementation, at least backed by Microsoft. I very much doubt it, as Microsoft uses deep learning for highly competitive products, such as Bing and Skype Translator: blogs.skype.com/2014/12/15/skype-translator-how-it-works. $\endgroup$ – Aleksandr Blekh Dec 17 '14 at 14:09
  • $\begingroup$ indeed....there is no open source DCN implementation available...!!! $\endgroup$ – hadooper Dec 18 '14 at 13:06
  • $\begingroup$ @hadooper: I was pretty sure of that. However, I am as well pretty sure that it is just a matter of time, until we will see some open source implementations of DCN, similarly to already existing neural networks and deep learning projects. $\endgroup$ – Aleksandr Blekh Dec 18 '14 at 13:36
  • 1
    $\begingroup$ Microsoft researchers achieved 0.83% test error rate with DCN and H2o is getting 0.87% error rate with deep learning for same mnist dataset...(SOURCE : slideshare.net/0xdata/h2-o-deeplearningarnocandel052114) $\endgroup$ – hadooper Dec 19 '14 at 5:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.