Monday, 2 November 2015

Anti-Virus Neural Network Should Catch More Malware

A small adjustment to malware can cause the virus threat is no longer detected, but an Israeli company claims to have found the solution: artificial neural networks. The neural networks to analyze the properties of millions of malware instances and clean files, and so learned to recognize the characteristics of malware.

The approach should ensure that the virus scanner of the company is better able to detect modified versions of malware that is missed by traditional anti-virus software. Deep learning, such as the approach is called, consists of training of a large network that consists of simulated neurons and synapses to get as complex patterns from the data provided. The intention is that the network can recognize the end of itself completely new ones.

The Israeli Deep Instinct says to train their own neural network with a large number of files and settings. A time- and computing-intensive process, which runs on a cluster of GPUs. According to founder and CTO Eli David can own the solution 20% more malware detection than existing solutions. So it can tell whether a file appears sufficient to existing malware makes it suspicious, says MIT Technology Review.

A small adjustment or particular string in the code is therefore no longer enough to deceive the virus. According to Professor George Cybenko the British Dartmouth College is the idea of ​​neural networks to find malware is not new, but gives the appearance of deep learning possible for renewed interest. He argues that the promised results must first be tested. In addition, malware authors are very persistent. "If there is a breakthrough, they will do research and come with a new approach," he warns.

No comments:

Post a Comment