Apple Developing 'Apple Neural Engine' Chip to Power AI in iOS Devices

[lincolntran]

having a better AI processor to process the data is a great first step.

But Apple's biggest problem remains *DATA*.

Without enough data, even the fastest & best AI chip - is still pretty much as bad as before


(their privacy policy is going to kill them/their future - because it's completely misapplied)

There are always more than one ways to do things.
[doublepost=1496037440][/doublepost]

They had ALL the datacenter horsepower you could imagine, and couldn't even get something as simple as Siri right... so they are supposingly implementing a chip into every iPhone?

The lack of information of this article is unprecedented.

You must be a genius. Siri might look simple but to me it's super complex by any measure.

 

[JamesPDX]

This notion of a Neural Engine is pure marketing and not of great concern -except your wallets or man-satchels. Murses?
Anyway, this is about jump-starting iPhone sales to where they were a few years ago.

You all know that we are many years away from true AGI, right? You'd better pray that we're at least 50 years away, because right now, the human race isn't ready for it. That's the new arms race. Apple will not win such a race, but they may bankrupt themselves by trying. Write it down.

 

[gnasher729]

having a better AI processor to process the data is a great first step.

But Apple's biggest problem remains *DATA*.

Without enough data, even the fastest & best AI chip - is still pretty much as bad as before

(their privacy policy is going to kill them/their future - because it's completely misapplied)

That's where I have to really disagree. You are more than welcome to send all your data to Google (which is apparently very interested now in gathering all your credit card information), and it may be a bit more convenient, but it means Google has all your information. I don't mind, actually - but I mind that they might have my information.
[doublepost=1496049160][/doublepost]

Apple is more concerned about privacy than improving Siri, which is wrong, If Siri has access to same data as Google Siri would be great, Siri development is hindered by Tim Cooks privacy first policy.

If Siri is hindered by "privacy first" then that's a good thing.

 

[Glockworkorange]

I would love better voice interaction. I can't believe Apple let themselves get so behind in this area in general. Can we say "slow turning ship?"

It's especially frustrating since Apple was a bit ahead of the game back in 2009 with the iPhone 3GS. That model introduced voice control. A little primitive, but you could see where they wanted to go. Can't recall whether voice control was handled on device. I think so. Anyway, I think they've been lapped by others at this point.

 

[GadgetBen]

Forget cars that drive themselves, just having the ability to use Siri to carry out simple tasks (without being connected to the internet) would be a start.

 

[emm386]

ou must be a genius. Siri might look simple but to me it's super complex by any measure.

whether I am a genius or not or Siri is complex or not is not the question.

Last time I checked, Apple was one of the richest companies on planet earth, with one of the largest data centers on earth. Taken into consideration, the computer technology is essentially their core business, the current state of Siri is - simply put - abominable.

Siri is - compared to competitors - lacking FAR behind in functionality, still sounds like a synthetic robot from the 90'S and even when asked for simple tasks, just offers me to "look it up on the Internet".

 

[kdarling]

It's especially frustrating since Apple was a bit ahead of the game back in 2009 with the iPhone 3GS. That model introduced voice control. A little primitive, but you could see where they wanted to go.

Historical note:

2009 was actually pretty late for voice control on a phone. By 2005 many flip phones had voice dialing by name or by digits, and smartphones had additional commands related to appointments, starting apps, playing music, etc.

For example, Microsoft had introduced speaker independent "Voice Command" for the Pocket PC and later Windows Mobile phones back around 2003. It worked well and got good reviews, such as this one.

(Others, like Nokia, had non-speaker-independent voice "tags", where you recorded and associated a specific phrase with a contact name or application. Which was kind of cool, because you could come up with funny/unexpected phrases that would call somebody or launch something.)

 

[iPadCary]

A "Neural Chip"? What is it, a brainiac or somethin'? tss tss

 

[Bacillus]

It's both. If Apple manages to make a powerful AI chip, that enables Siri to do the processing locally on the iPhone/iPad/Mac, then Siri doesn't have to send everything off to a server to be voice decoded there.

The problem is probably not contracting such a chip, but making a chip that doesn't deplete the iPhones battery in 10 minutes. I'm a developer, so I watched the sessions from last years WWDC (not only the keynote like most people). There was more detailed information on the AI image processing, that tags the images with metadata. When you take a photo with the iPhone, it makes more than 11 billion calculations to make this tagging, so that you can search for your cat and dog images, without having to tag the images manually. 11 billion calculations takes a lot of processing power. Powerful AI demands many heavy calculations like this, and that is why Siri currently sends the sound off to a server to be decoded, and Apple also collects data (anonymised) to make Siri better. I guess such information will still be sent to Apple, but Siri won't depend on sending to the server for decoding in the future. I'm sure it's be handled locally on the device.

So the device would explode with data ? It won't happen. Currently, it is even inconceivable that Siri would work without network, as its architecture is completely server-based.
If they want to change that paradigm, local decoding would obviously be better handled by a coporocessor - but it is the least of problems to be solved. While on-device tasks would require a limited instruction set, pre-recognition for local processing still requires syntax- and idiom decomposition which is soo data/register-intensive I don't see how that can't be handled locally.

 

[lincolntran]

whether I am a genius or not or Siri is complex or not is not the question.

Last time I checked, Apple was one of the richest companies on planet earth, with one of the largest data centers on earth. Taken into consideration, the computer technology is essentially their core business, the current state of Siri is - simply put - abominable.

Siri is - compared to competitors - lacking FAR behind in functionality, still sounds like a synthetic robot from the 90'S and even when asked for simple tasks, just offers me to "look it up on the Internet".

Can you give me some comparison data that I'm not aware of?

I've looked at youtube videos comparing Siri, Cortana, Google Now and they seem to be the same. There're both strength and weakness in all of them. They are all currently pretty basic and not really intelligent. To say that Siri is far behind is a little exaggerated and a misconception though.

 

[adnbek]

50% of the time....
Me: "Siri, remind me when I get home"

Siri: 'Okay, what do you want to be reminded about?'

In all fairness to Siri, if this was not in the context of a broader conversation you had with her, your request makes no sense and Siri is right to ask "about what?".

 

[afiza]

Hello, friends really this is a good one. what do you guys think about features of iPhone 7s plus? I have read one article i think that is a really a good one and more useful for everyone. Once look on the https://edgylabs.com/2017/07/06/latest-features-of-iphone-7s7s-plus-iphone-8-exposed/ and give your feedback through replying me or through twitter @afizabegum50.

 

[AxoNeuron]

I have been playing around a lot with machine learning in the new iOS 11 beta. I recently used Apple's new CoreML framework to import a handwriting recognition neural network I'd created in TensorFlow and it works brilliantly:

 

[thefredelement]

I have been playing around a lot with machine learning in the new iOS 11 beta. I recently used Apple's new CoreML framework to import a handwriting recognition neural network I'd created in TensorFlow and it works brilliantly:

This is awesome, great work and really fast recognition! Do you have this on git hub by chance? I'd love to take a look!

 

[chucker23n1]

Hello, friends really this is a good one. what do you guys think about features of iPhone 7s plus? I have read one article i think that is a really a good one and more useful for everyone. Once look on the LOL and give your feedback through replying me or through twitter @afizabegum50.

Well, for starters, we think spam is kinda uncool.

 

[AxoNeuron]

This is awesome, great work and really fast recognition! Do you have this on git hub by chance? I'd love to take a look!

The ML model file is a bit too big for Github so I've put it in a zip file here if you want to play around with it:

https://readycloud.netgear.com/client/dllink.html#t=0qoxux4l52m/Handwriting.zip

You need the Xcode 9 beta and iOS 11 beta if you want to test it on a real device. I wouldn't recommend putting the iOS 11 beta on your primary phone since it is quite buggy.

It's a very simple test project so don't expect it to recognize much beyond single, individual characters. It needs to be an upper-case alphanumeric character (A-Z, 0-9). You should draw the character as big as you can as well.

 

[thefredelement]

The ML model file is a bit too big for Github so I've put it in a zip file here if you want to play around with it:

https://readycloud.netgear.com/client/dllink.html#t=0qoxux4l52m/Handwriting.zip

You need the Xcode 9 beta and iOS 11 beta if you want to test it on a real device. I wouldn't recommend putting the iOS 11 beta on your primary phone since it is quite buggy.

It's a very simple test project so don't expect it to recognize much beyond single, individual characters. It needs to be an upper-case alphanumeric character (A-Z, 0-9). You should draw the character as big as you can as well.

Oh I get it all, thank you so much... I've been slammed at work but just finished up a project early this am, I've been dying to check out CoreML and ARKit.. I've snuck in a few lines into NLP and was really impressed.

The future looks incredible. I can't stop thinking about more advanced cameras and biometric sensors that can combine with ARKit and either show a view of like things behind walls or molecular compositions or if there are bad guys hiding, etc. Or also maybe an iDevice with some kind of emitter where we can combine this stuff and make holograms!

Sorry for my unbridled enthusiasm, I'm a work-a-day-dev and this stuff is crazy exciting.