Technology : Malicious viruses caught in the Net

2019-02-27 03:07:07

By Mark Ward COMPUTER viruses thrive on the Internet because it allows them to spread with ease. Every time a file or document is downloaded there is a chance a virus will be hitching a ride. Some viruses even hide in e-mail attachments and launch their attack when you read your messages. Now researchers at IBM are working on a way to block these destructive programs. “In the brave new world of the Internet, viruses can spread with extraordinary rapidity,” says Steve White, a senior virus researcher at IBM’s Thomas J. Watson laboratory in New York. And White believes there is only one way to stay on top of the problem: “We have to take humans out of the loop.” White and his colleague Gregory Sorkin have developed a neural network that will analyse computer viruses. Previous attempts to use neural networks failed, says Sorkin, because there were so few examples for them to learn from. While there are over 6000 known computer viruses they divide into a large number of distinct categories, each with only a few members. Typically neural networks need to be trained using hundreds of similar examples. To get around this problem, White and Sorkin “shredded” known computer viruses into 3-byte fragments. They trained the networks to recognise the common data structures, tasks and combinations of instructions shared by fragments from different species. This approach also meant that a virus was unlikely to be triggered when the neural net was sampling it, because the suspect program was never run as a whole. The result has already surprised White’s professional virus hunter colleagues. “The neural network found traits that were indicative of a virus that they had not thought of,