The Android app store, the Play Store, has several million applications. Sorting the bad guys' bad guys in terms of safety or privacy is no easy task. In a blog post published Wednesday, July 12, 2017, Google explains its methodology to flush out what it calls PHA, for Potentially Harmful Apps (potentially harmful applications).
Today, Google is among the leaders in artificial intelligence. You can find it in many of its products: from its search engine to its application Photos for Android, via Gmail and Google Translation. Not surprisingly, the company also uses this technology, deep learning to be precise, to help it detect these PHAs.
Concretely, Google compares automated applications of the same kind to each other to see if they do not require too many permissions over others. The company gives the example of a digital coloring book that would require access to the GPS of the user. As most coloring applications do not require this kind of permissions, this behavior will be judged as suspicious. It will then be up to a Google team to determine the sequence of events for that application.
AI to the rescue
Except that this procedure requires the creation of categories of applications rather fine and especially dynamic. Indeed, programs evolve rapidly today and the categories can not therefore remain fixed. This is where deep learning takes place. Google uses an algorithm that will analyze millions of Play Store applications, as well as their metadata such as text descriptions, requested permissions, metrics provided by users ... By learning from this mass of information, Algorithm can be very precise groups.
Thus, Google claims a rate of infection of users by malware via the Play Store at only 0.05% in 2016, a decrease of 0.15% compared to the previous year. Nevertheless, the system is far from perfect. Lookout mobile antivirus alone released 260 withdrawals of malware from the Play Store in 2016, a sign that some still manage to get through the cracks.
No comments:
Post a Comment