If you can read this, either the style sheet didn't load or you have an older browser that doesn't support style sheets. Try clearing your browser cache and refreshing the page.

(BBC-US)   That stealth bomber is a dog. I can tell by the pixels   ( bbc.com) divider line
    More: Awkward, Neural network, image recognition systems, Artificial intelligence, Neuroscience, Artificial neural network, adversarial examples, Networks, Unsupervised learning  
•       •       •

3229 clicks; posted to Geek » on 03 Nov 2017 at 5:50 PM (28 weeks ago)   |   Favorite    |   share:  Share on Twitter share via Email Share on Facebook   more»



21 Comments     (+0 »)
 
View Voting Results: Smartest and Funniest
 
2017-11-03 01:19:21 PM  
Please continue fearing our pending AI overlords. That field isn't 90% vapourware. Honest.
 
2017-11-03 04:37:29 PM  
img.fark.netView Full Size
 
2017-11-03 06:26:40 PM  
Humans can't tell if a dress is blue or gold, so I'm not going to be too critical.
 
2017-11-03 06:36:33 PM  

Ivo Shandor: [img.fark.net image 410x210]


There's a dog in the next apartment that I wouldn't mind giving flying lessons to.
 
2017-11-03 06:52:16 PM  
I, for one, welcome our new Al overlord:
indiepundit.comView Full Size



wait...what?
 
2017-11-03 06:54:59 PM  
That was supposed to be a response to the erster Pfosten.
 
2017-11-03 07:16:27 PM  
Nah, the AI was just trying to make an F-35 joke.
 
2017-11-03 07:34:56 PM  

RoomFullOfMonkeys: Humans can't tell if a dress is blue or gold, so I'm not going to be too critical.


The correct answer was "ugly."

Mr. Flare Explains: What Color The Dress Is
Youtube bg41XfnIBvk
 
2017-11-03 07:38:33 PM  

Tr0mBoNe: Please continue fearing our pending AI overlords. That field isn't 90% vapourware. Honest.


To be fair, I read about this elsewhere and looked at the sample pictures, and one of them that was supposed to be a dog and was misclassified by changing one pixel, looked nothing like a dog to me.

So I can't really criticize the AI if I can't identify stuff at all.
 
2017-11-03 07:42:48 PM  
And yet no pictures of dogs or bombers in TFA.

But I like turtles.
 
2017-11-03 08:16:01 PM  
 There's weaknesses now, but things will continue to improve. And, in most AI applications, it won't be a still image without context but video where some elements of context that can be employed. Humans have troubles with still images or tightly controlled videos (optical illusions, for example). It's improper to demand computer algorithms perform better.

So, relax. We'll all still get an opportunity to die at the wrong end of a murderbot when the time comes.
 
2017-11-03 08:51:17 PM  

Bondith: Ivo Shandor: [img.fark.net image 410x210]

There's a dog in the next apartment that I wouldn't mind giving flying lessons to.


This guy can give you some tips.
img.fark.netView Full Size
 
2017-11-03 09:46:41 PM  
Dear god, I hope that AI hasn't unlocked the secrets of basselope technology. For that would be the ultimate weapon in the lead up to Skynet!

img.fark.netView Full Size


img.fark.netView Full Size
 
2017-11-03 09:55:44 PM  
reading this article's headline, i was really looking forward to seeing the image before and after that one pixel was changed.  a picture, or two, really would have been worth a thousand words helping me understand what deep neural AI networked image recognition is about.

the fact it's missing tells me the writer has either misconstrued a point made in the interview or misrepresented it deliberately just to make way for a clickbait headline.
 
2017-11-03 09:59:00 PM  

American Decency Association: reading this article's headline, i was really looking forward to seeing the image before and after that one pixel was changed.  a picture, or two, really would have been worth a thousand words helping me understand what deep neural AI networked image recognition is about.


One pixel attack for fooling deep neural networks
 
2017-11-03 10:09:30 PM  

Ivo Shandor: American Decency Association: reading this article's headline, i was really looking forward to seeing the image before and after that one pixel was changed.  a picture, or two, really would have been worth a thousand words helping me understand what deep neural AI networked image recognition is about.

One pixel attack for fooling deep neural networks


im glad the the author otfa wasn't going for a cheap shot, but the lack of a simple example is even more frustrating after reading the author's apparent confidence quantifying the success of their own research
 
2017-11-04 12:09:06 AM  
Well it was well known in the '90's that it was only invisible to AMERICAN radar, so yah another boonDOGgle
 
2017-11-04 06:10:47 AM  
This is part of the problem I have with fully automated systems. If it works 99.9 non-terminating decimal % of the time, then it sometimes fails. When humans fail, there is a legal system to hold them accountable (I know, don't laugh). If a computer scans my brain and says "suicidal" even though I'm not, and I get removed from my job and put in a hospital, what recourse do I have?
 
2017-11-04 09:22:01 AM  

casual disregard: This is part of the problem I have with fully automated systems. If it works 99.9 non-terminating decimal % of the time, then it sometimes fails. When humans fail, there is a legal system to hold them accountable (I know, don't laugh). If a computer scans my brain and says "suicidal" even though I'm not, and I get removed from my job and put in a hospital, what recourse do I have?


99.999999...etc = 100.0
 
2017-11-04 11:14:29 AM  
Kinda cracks me up when there is free software that will identify duplicate and near duplicate photos.  Granted, that's a far cry from identifying what's in a picture.

This is truly awesome stuff and doesn't come with any PUPs or other bloatware that I could find.  And you don't even need to install it, per se, just download and run.  I was incredibly impressed at how well it works.

http://www.duplicate-finder.com/photo.​html

/If you have a brazillian copies of photos on your hard drive you'll thank me
 
2017-11-04 03:40:32 PM  
img.fark.netView Full Size

img.fark.netView Full Size

"Drop the rifle, you have five seconds to comply"
 
Displayed 21 of 21 comments

View Voting Results: Smartest and Funniest

This thread is archived, and closed to new comments.

Continue Farking

On Twitter





Top Commented
Javascript is required to view headlines in widget.
  1. Links are submitted by members of the Fark community.

  2. When community members submit a link, they also write a custom headline for the story.

  3. Other Farkers comment on the links. This is the number of comments. Click here to read them.

  4. Click here to submit a link.

Report