DV Hardware - bringing you the hottest news about processors, graphics cards, Intel, AMD, NVIDIA, hardware and technology!
   Home | News submit | News Archives | Reviews | Articles | Howto's | Advertise
 
DarkVision Hardware - Daily tech news
August 4, 2020 
Main Menu
Home
Info
News archives
Articles
Howto
Reviews
 

Who's Online
There are currently 95 people online.

 

Latest Reviews
Ewin Racing Flash gaming chair
Arctic BioniX F120 and F140 fans
Jaybird Freedom 2 wireless sport headphones
Ewin Racing Champion gaming chair
Zowie P-TF Rough mousepad
Zowie FK mouse
BitFenix Ronin case
Ozone Rage ST headset
 

Follow us
RSS
 

One pixel can be enough to fool AI image recognition

Posted on Monday, October 30 2017 @ 10:38:28 CET by


A group of researchers from the Japanese Kyushu University discovered AI-based image recognition tools can be really dumb and easily exploited. By altering a relatively small number of pixels, which would not be noticeable to a human observer, the AI-based systems can be tricked into classifying a photo of a car as a dog, or a dog as a cat:
As explained in a paper, the researchers came up with the startling conclusion that a one-pixel attack worked on nearly three-quarters of standard training images.

Not only that, but the boffins didn't need to know anything about the inside of the DDN – as they put it, they only needed its “black box” output of probability labels to function.

The attack was based on a technique called “differential evolution” (DE), an optimisation method which in this case identified the best target for their attack (the paper tested attacks against one, three, and five pixels).
INTC logo

By changing five pixels in a 1024 pixels photo, the researchers achieved a success rate of 87.3 percent. The big caveat here is that they used very small images. On a photo of 280,000 pixels, which is about 530 x 530 pixels, it would require the alteration of 273 pixels, which is still relatively little. Full details at The Register.



 



 

DV Hardware - Privacy statement
All logos and trademarks are property of their respective owner.
The comments are property of their posters, all the rest © 2002-2019 DM Media Group bvba