HomeНаука и техникаRelated VideosMore From: O'Reilly

OSCON 2013: Jeff Hawkins, "On Open Intelligence"

167 ratings | 12172 views
http://www.oscon.com Numenta is working on machine intelligence. Our approach is to first understand, in detail, how the neocortex works and then build systems that work on those principles. In the past few years we made good progress on both the theory and implementing the theory in software. The resulting technology is a type of memory system based on sparse representations and distributed sequence memory. Like the brain, systems built on this technology need to be trained on streaming sensory data. The system learns the spatial and temporal structure in sensory data. We have applied this technology to a commercial product called Grok. Grok is fed data from sensors, it discovers the structure in the data and uses this to make predictions and detect anomalies. Grok is being applied in diverse industries such as energy pricing, product forecasting, predictive maintenance, and finance. The technology we have created represents a new approach to machine learning and we believe it will be the fundamental component of all machine intelligence. We published the details of the theory in a white paper in 2009. There is substantial interest in working with this technology from both academics and technologists. In my talk I will describe our vision for machine intelligence. I will give a brief introduction to the technology. And I will describe why we are creating an open source project. http://www.oscon.com/oscon2013/public/schedule/detail/30583
Html code for embedding videos on your blog
Text Comments (19)
Kriszzzful (3 years ago)
Is this fast forwarded?
Grandfather_Din_Racket (4 years ago)
I think that Hawkins is on some portions of the right track (90% right). I think the principles he's discovered are all correct, but I think he's wrong about not wanting to build massively-feedbacked "tactile" robot bodies to go with his brains. (Or, he's even smarter than I thought, because he doesn't want the Luddites to rebel against his creations, and is simply in subterfuge / stealth mode.) I think that there will be superhumanly intelligent machines that improve their own bodies, much as human doctors and regenerative medicine experts have improved processes that will be used to help themselves. I think that putting a humanoid robot body on a Hawkins type brain will help that brain to model itself as human, when it engages in feedback with the world and experiences the world. This will also help it to identify with humans, if sufficient care is taken in its design and construction. The robot body should be allowed a full lab and a large budget, and be raised with kindness, starting off in a "slightly less powerful than human" child body. These brilliant children could then learn far faster than normal humans, and help assist humans in their adaptation to their environment (to increase robustness, intelligence, self-defense against sociopathic human networks AKA government, and longevity).
William L. Weaver (4 years ago)
*Jeff Hawkins Knows Brains* Leveraging the new Google+ Comment Integration on YouTube to Spread the Word. I've been following Jeff's work for the past 10 years. Boy, they sure have made a boat-load of progress. And now with NuPIC, advancement of the algorithms is an Open Source Project!
jacob man (4 years ago)
Ray Kurzweil is doing the same thing at Google, he's trying to get search engines to understand natural language by principals of the neocortex.
KiloSierraAlpha (5 years ago)
Yep, sorry :) I work in machine learning field and we all just shorten it since we write it all the time.
Srgg Cleftsow (5 years ago)
ML here is Machine Learning isn't it (I'm not sure about my desifer about that abbreviation is correct)?
KiloSierraAlpha (5 years ago)
Excellent. I'm glad that you think it's great. Publish a paper please :)
KiloSierraAlpha (5 years ago)
No, I don't agree at all. You're comparing something that can be demonstrably tested by anyone (i.e. the directions or the length of trip on Google or other maps) and claims that Grok is somehow awesome because it uses these "magical" cortical algorithms which are never compared against other ML algos with known properties.
Srgg Cleftsow (5 years ago)
Do you use google maps or google itself for example? I suppose you do. How much studies that are behind that technologies did you see? I mean that are different things -- studies and a final product. Wether it is successful or not depends of its commercial characteristics not of studies of subject. Don't you agree?
Tristan Slominski (5 years ago)
Fair enough, I confused your call for papers with call for source code.
KiloSierraAlpha (5 years ago)
Well, would you rather pay for snake oil and something that's not peer-reviewed or would you pay for something that's published and proven and you know its performance characteristics? It's fine if they patent the heck out of Grok but publish some studies comparing it to standard techniques please! Let's see how great it really is.
KiloSierraAlpha (5 years ago)
I've seen NuPic source. And it's not really impressive. It's mediocre in almost all of the areas they tried to market it in. It's mediocre at OCR, speech recognition, pattern recognition etc. It's next to impossible for NuPic-based model to beat some of the basic ML and pattern recognition algos. This is why it failed. I just wish Jeff & Co. published papers comparing it to state of the art stuff years ago.
Tristan Slominski (5 years ago)
Checkout Github / numenta / nupic ... open source now, surprised me.
Srgg Cleftsow (5 years ago)
Why a commercial app must interest science people, not consumers?
KiloSierraAlpha (5 years ago)
Numenta still won't publish a peer-reviewed study of their Grok platform against some well-known ML techniques. Probably because it's mediocre like NuPic was. They're doing heavy marketing without much scientific study. This is why Numenta is not respected in scientific community (especially in ML community) and most researches just ignore Jeff's talks. Until they start releasing papers, safely chalk it down as hype.
Jessica Taylor (5 years ago)
I study neuroscience, and the cortex is generally 6 layers, and a minimum of 4 towards the edges of the cortex.
halooooooooom (5 years ago)
did he just say 'tooken'. ouch
sachamm (5 years ago)
Wow, great talk.
AN OUTLAW INLAW (5 years ago)
I am sure the military complex will be very interested.

Would you like to comment?

Join YouTube for a free account, or sign in if you are already a member.