With Google’s own open source machine learning framework TensorFlow, Gmail successfully used artificial intelligence to identify the most difficult spam types. Gmail, which learns from TF machines, now blocks an additional 100 million spam messages per day. Google has been able to block 99.9% of spam through rule-based filters, but the last 0.1% is the most difficult to identify.
Neil Kumaran, product manager for Google’s anti-technical abuse division, said
Where did we find these 100 million extra spam messages? We’re now blocking spam categories that used to be very hard to detect. Using TensorFlow has helped us block image-based messages, emails with hidden embedded content, and messages from newly created domains that try to hide a low volume of spammy messages within legitimate traffic.
For many years, Gmail has been using AI in addition to rule-based filters. Machine learning will learn to find hidden patterns in which emails cannot be trusted. Algorithms trained in this way measure a large number of metrics, from the format of the email to the time of delivery. Kumaran said that TensorFlow can manage this data more easily, and the open source nature of the framework means that new research from the community can be quickly integrated.