backtop


Print 32 comment(s) - last by mik123.. on Oct 15 at 3:20 AM


Qualcomm's NPU-powered robot  (Source: slashgear.com)
The NPUs could be used in artificial vision sensors and even smartphones

Qualcomm has created chips that are capable of learning about their environment and behaving accordingly, and they're ready for commercial development. 

According to it news, Qualcomm announced the new chips at the EmTech Massachusetts Institute of Technology conference this week. The idea is to use them in the medical field as well as mobile devices like smartphones. 

The chips -- called neural processing units (NPU) -- aim to work like a brain, where computer circuits imitate neural structures and have the ability to learn their environment and behave differently from these experiences. 

Qualcomm and its partner for the venture, Brain Corporation, have already created prototypes of the chips. They're now preparing for commercial design and development.

The NPUs have already been tested. They were used in a four-wheeled robot that can be trained to learn navigation using the commands "good robot" and "bad robot." 

In the future, the NPUs could be used in artificial vision sensors and even smartphones, where the devices learn their users' habits and act accordingly for greater convenience. 

Qualcomm chief technology officer Matt Grob said that his company and its partners will design and manufacture the NPUs by next year.

Qualcomm isn't the only company to do this. Back in August, IBM revealed the latest version of its neuron-based chips as part of its SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) program. 

Source: it news



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

''Learn'
By Wazza1234 on 10/12/2013 3:27:13 AM , Rating: 5
OK so Google has created a huge search engine server cluster which 'learns' what you wanted to search for... when you type it.

Facebook 'learns' whether I like a picture or not... when I click 'like'.

Light sensitive light switches 'learn' when it's dark or not... by using a light detector.

Seriously, when did we start using the word 'learn' when all these systems are doing is detecting the result of sensors or an input, exactly like any web system or program has done, ever.

It's like we're desperate to start labelling things 'artificial intelligence' when there's no intelligence there - certainly not self-formed intelligence, it's only doing or reading whatever we program it to.

And the same goes for any attempt thus far to create 'AI' - none of them qualify as far as I'm concerned.




RE: ''Learn'
By mr.mac52 on 10/12/2013 12:15:32 PM , Rating: 3
This is why I call AI Artificial Ignorance


RE: ''Learn'
By mik123 on 10/12/2013 3:33:18 PM , Rating: 2
How would you build it?


RE: ''Learn'
By Wazza1234 on 10/13/2013 2:20:53 AM , Rating: 2
I don't know, I don't think anyone knows.

The key is that we don't program the capability in, but whatever we create establishes its own intelligence and capabilities, or programs itself.

And that doesn't include those robots which can 'learn' to recognise objects etc - that's still just a computer program written to be informed about an object and store it.

It may well not be possible - there are some elements of existence, for example perception (where you see the world from) which haven't yet been explained by science.


RE: ''Learn'
By mik123 on 10/13/2013 4:02:04 AM , Rating: 2
So, what is the fundamental difference between (for example) a toddler learning to walk, and a neural network learning to control a robot?


RE: ''Learn'
By Wazza1234 on 10/14/2013 2:52:22 AM , Rating: 2
quote:
what is the fundamental difference between (for example) a toddler learning to walk, and a neural network learning to control a robot?


Are you talking about the rat neurons? If so, I would argue that isn't artificial. They've effectively grown a rat brain using rat cells, just like we can grow ears using stem cells. It's using nature to recreate nature. For me, it's not artificial intelligence.

The body is artificial, but we can do that with prosthetic legs.


RE: ''Learn'
By mik123 on 10/14/2013 4:06:06 PM , Rating: 2
What rat neurons? I'm talking about neural networks implemented in software.
Those network can learn to recognize images, or they can learn how to balance robot movements.


RE: ''Learn'
By Wazza1234 on 10/14/2013 4:06:49 PM , Rating: 2
No. They are not real neutral networks.

And they can only 'learn' to recognise images because they've been programmed to do so.


RE: ''Learn'
By mik123 on 10/15/2013 12:50:16 AM , Rating: 2
Babies learn to recognize images because they are "programmed" to do so in their DNA. They don't "decide" to learn, they are hardwired to do so by nature.

So, what is the difference?


RE: ''Learn'
By Wazza1234 on 10/15/2013 2:08:54 AM , Rating: 2
Don't even dare to try to pretend that anyone knows how real natural neural networks work.

ANN are an extremely crude type of computer model based on a very simplistic idea of how neurons work using an out of date theory.

The 'learning' methods that they employ are no more advanced than regression algorithms or statistical models which have been around for decades. Calling it artificial intelligence or learning is ridiculous.

quote:
They don't "decide" to learn


I think that's exactly what they do, and that's a crucial difference too. Computers can only learn what you program them to learn - the 'output' operators in the fake neural networks. Real brains can learn anything, if they choose to.


RE: ''Learn'
By Jaybus on 10/14/2013 1:14:02 PM , Rating: 2
It is not black and white, ie. we either know or we don't know. Artificial neural networks attempt to model biological neural networks (brains). Neural science does not know everything there is to know about animal brains, but does know quite a bit.

There is both programming and learning. In a brain, the starting state is programmed in DNA. In the ANN it is programmed by humans. For example, a baby doesn't learn to move a particular set of muscles to move the fingers and make a fist. The starting state for each neuron has the ability to fire an associated muscle given a set of synaptic inputs. What is learned is control of that motor function. Each neuron is fine tuned by adjusting the output for given sets of inputs. An ANN emulates this method of learning by adjusting output relative to input, keeping the changes if they produce better results and discarding the changes if they produce worse results.

It is a matter of complexity. A human brain has 100 billion neurons and trillions of synapses. Current ANNs have many orders of magnitude less resources. Biology is incredibly energy efficient. ANNs are extremely power constrained by comparison.

Fortunately, it doesn't have to be as smart as a human brain to be useful. Biological brains are performing countless tasks. We can dedicate an entire ANN to a single task, since it doesn't have to control life support functions, etc. So an ANN on the level of an insect brain could be extremely useful. Frog brain level is probably more than adequate to learn to safely drive a car.


RE: ''Learn'
By flyingpants1 on 10/12/2013 4:47:55 PM , Rating: 2
You misunderstood the way Google Now works


RE: ''Learn'
By Wazza1234 on 10/13/2013 2:18:40 AM , Rating: 3
Actually I was talking about Google Search - as in the Google home page.

The point being that if we were to describe 'receiving an input' and 'reacting to that input' as 'learning' - it would cover any system ever built.

As for Google Now, it's a more complicated algorithm but it's still a series of inputs being processed by an algorithm and a response determined. It's not artificial intelligence.


RE: ''Learn'
By Jaybus on 10/14/2013 1:43:05 PM , Rating: 2
An ANN works in a fundamentally different way. Google's algorithmic search sorts by the probability of a page matching an input. The probability itself is calculated in a deterministic way. Accuracy is dependent on the probability calculation. ANNs are not algorithmic in this way. Instead they use feedback to very simply adjust output relative to input by trial and error until it gives better answers. If that is not learning, then I'd like to know what is?


RE: ''Learn'
By Wazza1234 on 10/14/2013 4:09:19 PM , Rating: 2
quote:
ANNs are not algorithmic in this way. Instead they use feedback to very simply adjust output relative to input by trial and error until it gives better answers


If you don't understand that behind that is an algorithm which Google very deliberately programmed in, I can't help you. In fact, if you think any computer doing anything isn't following an algorithm, or to use a synonym, machine code, you're sadly mistaken.

And you're right to ask what 'learning' is - since no computer program qualifies. A computer which didn't have any programming to begin with but which learned how to answer things through its own development would be true learning, and that does not exist yet.


RE: ''Learn'
By mik123 on 10/15/2013 12:58:18 AM , Rating: 2
What makes you think a brain does not follow some kind of an algorithm, or a development blueprint encoded in DNA?


RE: ''Learn'
By Wazza1234 on 10/15/2013 2:19:03 AM , Rating: 2
Well let me put it this way. If I create a computer program, I don't believe I've created intelligence. And an algorithm is just that, a computer program. Many of the computer programs for ANN's simply employ formulae developed by statisticians centuries ago.

I see where the conversation is headed though - how complex / advanced does an algorithm have to get before it crosses the line from a program to intelligence. My honest opinion is that nobody knows, yet. We will build more complicated computer programs which will be able to answer any question, have normal conversations, will be possibly able to move and act as if it's intelligent, and everybody will think AI is real. But until we understand how our own intelligence works, it will always only ever do what it was programmed to do and it will only ever be the equivalent of a computer programming function taking a collection of inputs, processing them through a model, and producing an output.

I guess the thing for me which separates out the mind from any other object in the universe - is consciousness. It's the single biggest thing that nobody understands. If you can reduce a brain to simple chemistry, and take the deterministic perception on life, why would I personally be seeing the world at all, let alone from the perspective of my brain. I can create a light sensor in my workshop, it doesn't suddenly have a perspective on the world because it can detect light. But I do, my perception is somehow associated to the particular chemicals which form my brain. Why would I witness the world from the perspective of some chemicals? That makes no sense. That is the biggest mystery of science and I think the difference between a computer program and intelligence.

I accept it's a totally grey area but I'm just not yet willing to accept computer models as intelligence.


RE: ''Learn'
By mik123 on 10/15/2013 3:20:48 AM , Rating: 2
Your answer makes sense. In fact, this is how I used to think a few years ago. Then I came across a book "On Intelligence" by Jeff Hawkins, which explained a few things about a brain. If you're interested in this topic, I suggest you check it out.

Another question - where in evolution chain does consciousness start? Is it only present in us, modern humans? How about Neanderthals? Australopithecus? Chimps? Dolphins?
If you think about it, intelligence developed gradually from species to species. If you start with something simple, like a worm, you might be able to build its precise model on a computer - there will be mo mysteries left about that worm. Once you done that, move on to a more complex organism, such as an ant, then a frog, then a mouse, then a cat. Eventually you will be knocking on the door of "real" intelligence.


Breaking News
By venym76 on 10/11/2013 2:59:33 PM , Rating: 5
In other news Qualcomm will now go by their new name Cyberdyne Systems Corporation....




RE: Breaking News
By vision33r on 10/11/2013 3:14:09 PM , Rating: 2
They are far from that.. This is pure marketing. They better be ready because Intel is now 100% focused on mobility.


RE: Breaking News
By douchefree on 10/11/2013 9:06:44 PM , Rating: 4
While it's not at all clear to me what an npu is, at the logic level, they are far from a fluff firm.
Qualcomm is as large as Intel, and have vast amounts of experience in the mobile sector.


RE: Breaking News
By augiem on 10/12/2013 2:53:55 PM , Rating: 2
quote:
Qualcomm is as large as Intel,


Intel:
Revenue: US$ 53.34 billion (2012)
Operating income: US$ 14.63 billion (2012)
Net income: US$ 11.00 billion (2012)
Total assets: US$ 84.35 billion (2012)
Total equity: US$ 51.20 billion (2012)
Employees: 104,700 (2012)

Qualcomm:
Revenue: US$ 19.12 billion (2012)
Operating income: US$ 5.59 billion (2012)
Profit: US$ 6.10 billion (2012)
Total assets: US$ 43.01 billion (2012)
Total equity: US$ 33.52 billion (2012)
Employees: 26,000 (2012)

Depending on what your measurement is, Intel is 2x-4x bigger than Qualcomm.

Sources:
http://en.wikipedia.org/wiki/Qualcomm
http://en.wikipedia.org/wiki/Intel


RE: Breaking News
By Connoisseur on 10/11/2013 3:27:43 PM , Rating: 4
"My CPU is a neural net processor... a learning computer" - ::cue ominous T2 Theme Song"


RE: Breaking News
By YearOfTheDingo on 10/11/2013 3:44:50 PM , Rating: 5
Don't worry, we can retrain the Terminate to fight on our side with the commands "good robot" and "bad robot."


RE: Breaking News
By SilthDraeth on 10/12/2013 10:23:43 PM , Rating: 2
Give this guy a 6?


RE: Breaking News
By jtemplin on 10/11/2013 4:40:12 PM , Rating: 3
"Who said you could eat MY cookies!?"


RE: Breaking News
By Alexvrb on 10/13/2013 12:44:02 AM , Rating: 2
"Put the cookie down! NOW!"


RE: Breaking News
By RedemptionAD on 10/13/2013 1:53:32 PM , Rating: 2
Good Robot!!


RE: Breaking News
By DiscoWade on 10/11/2013 4:00:40 PM , Rating: 3
I, for one, welcome our new robot overlords.


RE: Breaking News
By venym76 on 10/11/2013 5:09:37 PM , Rating: 3
They couldn't screw up more than our Human ones.


RE: Breaking News
By ranran on 10/11/2013 5:56:51 PM , Rating: 2
Lol..you guys are too funny, love the comments on this story.... :) Thanks for ending my Friday on a laughing note.... :)


Cool
By Stuka on 10/13/2013 10:20:35 AM , Rating: 2
Maybe they'll make enough money off this that they can pay HTC's licensing fees to Nokia for the patents that Qualcomm has infringed on.




"We basically took a look at this situation and said, this is bullshit." -- Newegg Chief Legal Officer Lee Cheng's take on patent troll Soverain














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki