Undesired emails (spam) are responsible for the loss of more than $70 billion annually in the U.S. alone. Current-day spam filters work on sets of rules that are semi-manually set by humans which apply to both the single email metadata (e.g. sender address, format) and content (e.g. text body, attachments). Since the cost of creating and sending millions of emails with small variations among them is increasingly cheaper, finding and adapting to these rules is quite easy for a spammer, thus requiring continuous tuning and adding of new rules by the spam filter companies in an endless cat-and- mouse game.
“By leveraging the power of neural networks and machine learning, we are able to deliver a prototype solution that was trained to identify spam email without requiring a person to create a spam identification rule."
Professor Rajesh Vasa - DSTIL
The collaboration with MailGuard was to create a new continuously learning platform to detect spam. A set of deep neural networks and machine learning algorithms will work in parallel with MailGuard’s existing rule-based architecture to teach a machine how to filter spam emails. The aim of the new approach is to reduce the overall human effort required to detect spam.