Google says it's blocking 100 million more spam messages a day in Gmail after advancing its machine learning models. Specifically, Google has used TensorFlow, its open-source machine learning framework, to more efficiently modify its spam detection features. That enables it to spot the hardest-to-detect spam messages, such as image-based messages.
Google was already using machine learning to power its spam detection capabilities. And the tech company says the existing models, in conjunction with other protections, helped block more than 99.9 percent of spam, phishing, and malware from reaching Gmail inboxes.
Still, spammers continue to refine their techniques. Even though spam email has been a problem for decades, it's actually become more prevalent in recent years, according to the security company F-Secure, as systems have become more secure against software exploits and vulnerabilities.
Machine learning algorithms can identify patterns in ever-evolving spam messages that humans may not catch. Using TensorFlow, Google can train and experiment with different machine learning models more efficiently. In addition to image-based spam messages, TensorFlow has helped Google detect emails with hidden embedded content and messages from newly created domains that try to hide a low volume of spammy messages within legitimate traffic.
While Google stresses that security is one of Gmail and G Suite's main selling points, security challenges still exist, of course. A newly-published report highlights how scammers are using Gmail's "dot accounts" -- a feature of Gmail addresses that ignores dot characters inside Gmail usernames -- for fraudulent activity.
Prior and related coverage: