Tech News

AI uses a lot of energy. Hackers can make it eat more.


Attack: But this type of color means that if you can change the content, such as the image that was fed, you can change the amount that you need to use in order to deal with it. This opens the door for potential fraud, as researchers at the Maryland Cybersecurity Center reported in a new paper submitted to International Conference on Learning this week. By adding less noise to the network connection, he made the graphics more complex and increased its readability.

Assuming that the fighter had sufficient knowledge of the neural networks, he was able to increase his strength. While assuming that the shooter had no experience, he was able to reduce the network performance and increase power consumption by 20 to 80%. The reason, as researchers have found, is that the threats flow across a number of different radio stations. Creating a single visual presentation is enough to confuse many, says Yiğitcan Kaya, a PhD student and co-author.

Plunder: This type of attack is still speculative. Adjustable architecture has not yet been used here. But the researchers believe that this will change rapidly from the company’s complexity to the use of light neural networks, such as smart home devices and other IoT devices. Tudor Dumitraş, a professor who co-authored the study, says that more work is needed to understand how such a threat can harm the environment. But, he adds, the paper is the first part of the public awareness campaign: “The important thing for me is to make people aware that this is a new form of intimidation, and this can happen.”


Source link

Related Articles

Leave a Reply

Back to top button