REGISTER | LOGIN
Breaking News
News & Analysis

Keynoter: Noise Analysis Beats Google Now

11/2/2017 01:01 PM EDT
10 comments
NO RATINGS
Page 1 / 3 Next >
View Comments: Newest First | Oldest First | Threaded View
R_Colin_Johnson
User Rank
Author
Re: Noise is the Signal
R_Colin_Johnson   11/5/2017 12:57:58 PM
NO RATINGS
Yes, the thing about good research is that it often does not make good products. No matter the reason Sony's beta video cartridges were better than VHS but only pros used it, likewise with thousand of other examples. The only consolation we have here is open source, which Intel promises to supply with these innovative uses of realtime sensors. But ultimately the market will decide good, bad, benevolent or evil.

name99
User Rank
Rookie
Re: Noise is the Signal
name99   11/5/2017 12:39:58 PM
NO RATINGS
The problem with this stuff is that Intel has no delivery vehicle for it. They've been beating this and equivalent drums (eg smart health care, biometrics) for YEARS, but it doesn't matter to the real world because Intel doesn't sell the devices in which these products naturally fit, and seems incapable of, or uninterested in, making the deals that would get them into those devices. 

Compare for example the limited takeup of True Key/RealSense with the excitement around Apple's FaceID.

Ultimately it looks like it will play out the same as Xerox's UI work (we know how that turned out) and MS Research's work on how to make a simple secure OS (which informed a lot of iOS). Good useful research --- which the parent company doesn't have the vision to do anything with...

realjjj
User Rank
CEO
Re: ....
realjjj   11/4/2017 12:32:53 AM
NO RATINGS
Hope you had a good flight.

Some memory of the data will be required, you want your glasses or robot to know you, not humans in general. With only real time data it would be like talking to a stranger instead of an old friend. In some situations it is also less compute intensive, you don't want your robot to 3D map everything and categorize every object 240 times per second (LOL), it will do it once and then track changes- just like we do.

The way users are forced to give up data today is nuts, less so with Google Now as the data is fundamental for the service but everybody just strong-arms users into giving away their data. Ofc Google gets so much exposure that they have to handle the data with more care than most, many others simply sell it to anyone willing to pay for it. The good news is that the edge is gaining in importance and our future devices will handle most of the data locally. The way I look at it, the guys at the edge and their suppliers will want to create value and generate revenue by doing it at the edge.

What Intel is doing is interesting, could be more efficient and/or enable lower costs, could add some redundancy. Plus it's creative and kinda fun. How they try to bring it to market will be very important, humans don't like complexity, we always try to simplify everything. If they focus on lowering costs or creating new sensors it might be best in the short term. Otherwise folks might look at it as excessive, too convoluted and simply not worth the trouble. I like the idea of new sensors best for the next 5 years or so. What data to collect and why, how to collect it. Most sensors are single purpose today and that's not efficient. I suppose Intel would also be interested in an "awareness processor" (fancy name for a sensor hub).

R_Colin_Johnson
User Rank
Author
Re: ....
R_Colin_Johnson   11/3/2017 7:09:18 AM
NO RATINGS
You make some really good points, and thanks for taking the time to think this thing through. By the way, I'm not in favor of either side, just reporting the news (quickly). That said, you are right--Google can just watch what Intel does and then add that to its database of tricks, including your growing personal dossier. For instance, in the food burning example, Google is going to file that fact away--that you are a burner, that you are particularly bad at french fries and maybe even givie you a stream of ads (the next time you surf) about self-cooking french fryers that always come out perfect! What it boils down to is whether you think dossiers are intrusive, whether your equipment will still serve you well after Mr. Robot wipes your database and whether using sensors in novel ways in which they were not intended to be used is fun and innovative or an open-door waiting for hackers to use (once again) to outsmart the cybercops. That's all I've got at 4am in San Jose and you may be a clearer thinker at those hours, but I have to pack for the long flight back to Florida and I really really do appreciate your careful analysis of my story. Thanks realjjj.

realjjj
User Rank
CEO
Re: ....
realjjj   11/3/2017 4:32:44 AM
NO RATINGS
If you think about it, they don't quite frame it right at all. This solution doesn't have to be limited to real time data, and ofc Google can take advantage of real time data too.

You can't have personalized and in depth knowledge with only real time data.  But this is not what Intel's solution is about, real time vs in depth. It is about how you acquire the data and how to acquire more data so if anything, Intell's solution is more data intensive. The fact that they focus on real time doesn't mean that this is limited to real time and it would be counterproductive to limit it that way. They limit waste, they extract more gold out of the same amount of dirt but that has nothing to do with the next steps, you react to it and you either discard it or you create some memory of it - that's a choice available to you no matter how the data is collected.

The Google claim is gratuitous.

Edit: Let's take the burning food example. If you have a user that likes to burn french fries but nothing else, Google will use a device with machine vision to monitor the food while Intel will use the phone's mic to monitor the noise made by the air purifier and figure out when the food is burning. Both Google and Intel can also figure out what the user is cooking but they need to store some data to figure out that the user likes to burn french fries.

R_Colin_Johnson
User Rank
Author
Re: ....
R_Colin_Johnson   11/3/2017 3:27:27 AM
NO RATINGS
I'm sorry you got the impression that Intel was attacking Google, whereas in fact they were saying that Google Now is the BEST of the lot, but that they could do better without have to keep a dossier on their users.

R_Colin_Johnson
User Rank
Author
Re: Noise is the Signal
R_Colin_Johnson   11/3/2017 3:24:46 AM
NO RATINGS
I agree about Intel out-Samsunging Sansung, unfortunately I felt constrained to report only the facts: namely that Intel's Keynote Speaker said she admired Google Now as the best at giving ad hoc advise, but thought her group could do even better without the need to compile a secret dossier on their users. Thanks again for your excellent observation. We always love to hear our readers opinions!

realjjj
User Rank
CEO
....
realjjj   11/3/2017 12:23:52 AM
NO RATINGS
Very interesting but they should not attack Google, there is no need for that. Ideally a hybrid solution is best (the data could stay on device though), the two approaches are complementary.

Deep learning is mostly about sensors today and making the most out of the sensor data seems like a great idea. The increased complexity might be a bit scary though. Maybe the right play is to do more with fewer sensors. Glasses for example will require too many sensors given how restrictive the form factor is and using a few basic multipurpose sensors could be a solution.Maybe this can even inspire new types of sensors.

One more tiny thing, it's important to make a distinction between what we want and what we need.

photonic
User Rank
Author
Re: Noise is the Signal
photonic   11/2/2017 11:40:45 PM
NO RATINGS
A column about how Intel plans to out-Samsung Samsung would be much more to the point.

R_Colin_Johnson
User Rank
Author
Noise is the Signal
R_Colin_Johnson   11/2/2017 10:27:49 PM
NO RATINGS
Intel claims that sensor noise can be used as the signal to out-Google Google Now, Siri, Alexa and the rest. Intel claims to beating Google at its own game by repurposing the use of smartphone and Internet of Things sensors enables making useful realtime ad hoc suggestions to users without keeping secret dossiers on them.

Like Us on Facebook
EE Times on Twitter
EE Times Twitter Feed