Fake AI scanners can come back to reveal their faces they have been taught

However this assumes that you can get the most out of these courses, says Kautz. He and his co-workers at Nvidia came up with another way to reveal their secrets, including facial expressions and other things, more medical and more, which does not require the opportunity to get an education at all.

Instead, they developed strategies that would bring back what they had been taught by the trained type changing the paths that the race goes through in fixing the same. Take a well-known visual network: to identify the content of the image, the network passes through several types of artificial neurons, and each section releases a variety of information components, from the sides, to the shape, to other known objects.

Kautz’s team found that they were able to distort the color between the steps and change the direction they were directing, in order to restore the image captured from the content of the color. They tested this method on a variety of well-known images and GANs. In one experiment, he showed that he could accurately reproduce images from ImageNet, one of the most well-known pages with images.

ImageNet images (above) in addition to retouching images created and retrieving the trained image at ImageNet (below)


Like Webster’s work, the modified images are similar to the actual ones. “We were amazed by the final race,” says Kautz.

Researchers say that this type of attack is not the only myth. Phones and other small devices are starting to use more AI. Due to battery and memory problems, some models are sometimes processed half of the device and sent to the cloud for final computing, a process called split computing. Many researchers think that computer hacking will not reveal any secrets from a person’s cell phone because it is the only type that is shared, says Kautz. But his attack proves otherwise.

Kautz and his colleagues are now working to find ways to prevent racism from revealing their secrets. We wanted to understand the risks in order to reduce the risk, he says.

Although he uses a variety of methods, he thinks his work with Webster works well together. Webster’s team pointed out that confidential information can be obtained by releasing a color; Kautz’s team indicated that their personal information could be disclosed in retrospect, and retrieved the content. “Investigating both sides is important to better understand how to avoid the risks,” says Kautz.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *