Numenta – Neural Networks the Next Generation

Cover of "On Intelligence"
Cover of On Intelligence

I am rapidly falling behind on machine learning technology. Numenta is a company that is working on Hierarchical Temporal Memory, or HTM. I first learned about HTMs in the book On Intelligence by Numenta co-founder Jeff Hawkins (who was also a major player at a little company called Palm). Jeff left the software world and delved into deep brain research, trying to better understand human intelligence. Now he has taken what he learned and turned it back into software – the HTM.

Like I said, I am a little behind in recent advances, so I might be ostracized if I call HTMs a form of neural network. I will take my chances for now. I am on the Numenta mail list and just got this email – if you are interested in the NuPIC software, go and download it for free.

We now have available a new release of NuPIC and the Vision Toolkit, improvements to the Web Services user interface, plus a new People Tracking demo application. There are a few items of note in this new release.

First, we have added some super functionality to the Vision Toolkit that allows you to search the web for images and to automatically download hundreds at a time into the toolkit. Populating your network with images has proven to be the hardest part of using the Vision Toolkit, so we are pleased to be able to dramatically accelerate that task. You can experiment with this by selecting “Import” from the “Prepare Images” screen. The new release of the Vision Toolkit also fixes most of the bugs that users have reported.

Second, NuPIC 1.7.1 is now available. In addition to bug fixes, we have added an example that shows you how to convert from the Vision Toolkit to NuPIC. If you are interested in working on a vision problem, you now can start easily in the Vision Toolkit, but migrate to NuPIC when you need more functionality.

Third, we’ve made some nice improvements to the Numenta Web Services user interface, as well as added sample images for each of the networks. The sample images allow you to experiment with the networks without having to find your own images. We’ve tried to include images that illustrate the range of variations that are possible for each network.

Finally, we have included a new demonstration application. The People Tracking demo illustrates how HTM can be applied to video recognition. The HTM network in this demo has been trained to locate people in surveillance video footage. You can see how it operates by experimenting with the sample videos that are bundled with this demo. This network was developed in partnership with Vitamin D for their upcoming Smart Video application.

We’ve seen lots of interesting experimental Vision Toolkit networks created by the developer community. Feel free to be in touch withus if you would like to discuss commercial applications, to push
the Vision Toolkit into new areas of exploration, or to apply our technology to video recognition applications.

Remember that I am now posting occasionally on Twitter. Sign up as a follower of ddubinsky if you’d like to be alerted to interesting news from Numenta.

Finally, we have internships available!

Reblog this post [with Zemanta]