Profile cover photo
Profile photo
Rafael Barbolo
191 followers -
Partner at Infosimples
Partner at Infosimples

191 followers
About
Posts

A question for the community: do you think it's possible to build a general image classifier almost as good as those from Cloud APIs (Google, IBM, Microsoft, Clarifai, Amazon, Baidu ...) using only open source models and datasets?

I think it may be possible to get a pretty good classifier using transfer learning from pre-trained models they have shared themselves. If that is the case, then isn't being so open about their models a threat to their businesses?

What do you think about this discussion?

Post has attachment

Post has attachment

Post has attachment
I'm not sure if this is recent news, but Hinton is going to start a new Coursera class on NN on September:

https://www.coursera.org/learn/neural-networks

Post has attachment
Rafael Barbolo commented on a post on Blogger.
I'm eager to access the Cloud ML preview. Is it possible to process my limited preview request any faster?

Post has shared content
New Deep Speech 2 results, with lots of technical details! Even though English and Mandarin are vastly different, the same architecture does well on both; it can also learn a bilingual model. We also have a new architecture, called Batch Dispatch, to use GPUs not just for training but for serving models. Paper also talks about the Sortagrad algorithm; our HPC architecture; how to get BatchNorm to work on RNNs; and more. http://arxiv.org/abs/1512.02595

Post has attachment

Post has attachment
Wait while more posts are being loaded