Google can quickly modify its ML models to find the most difficult-to-detect spam messages with the open-source machine learning framework.
Google says that after advancing its machine learning models, it blocks 100 million more spam messages a day in Gmail. In particular, Google has used TensorFlow, its open source machine learning framework, to modify its spam detection functions more efficiently.
This allows it to detect spam messages that are hardest to detect, such as image-based messages. Google used machine learning already to power its spam detection capabilities. And the tech company says the existing models helped to block more than 99.9 percent of spam, phishing and malware from reaching Gmail inboxes in conjunction with other protections.
Spammers are still refining their techniques. Although spam email has been a problem for decades, it has become more prevalent in recent years, according to the security company F-Secure, as software exploits and vulnerabilities have become more secure.
Algorithms for machine learning can identify patterns in spam messages that people can not catch. Google can train and experiment more efficiently with different machine learning models using TensorFlow. In addition to image-based spam messages, TensorFlow has helped Google detect emails containing hidden embedded content and messages from newly created domains that attempt to hide a low volume of spam messages in legitimate traffic.
While Google emphasizes that security is one of the main sales points for Gmail and G Suite, there are of course still security challenges. A newly published report highlights how scammers use “dot accounts “from Gmail-a feature of Gmail addresses that ignore dot characters within Gmail usernames-for fraudulent activity.
