Shulou(Shulou.com)05/31 Report--

Did you know that employees only spend 5% of their time brainstorming creatively? According to Google's research, this is true. Most of the tasks such as tracking, research and analysis, final retouching, formatting and integrated feedback are done through email. In fact, the average person spends 11 hours a week filtering work messages, and about 14.5 million spam messages are sent (or received) every day, which only makes people more distracted.

Fortunately, for the 1.5 billion people who use Gmail and the 5 million employees of Gmail to G suite each month, the good news is that Google is watching you. As early as 2015, it claimed that 99.9% of spam, phishing and malware helped reach the account artificial neural network (the mathematical function layer loosely mimics brain neurons), while ensuring the sharing of legitimate emails. inadvertently end up in the spam folder below 0.05%. (in 2012, the proportion of spam was 99%. Now Google says it can block about 100m additional spam messages a day, in part because of the "new protection" provided by its open source machine learning framework, TensorFlow.

The company has previously said it protects users from 10 million spam and malicious emails every minute.

"We used to use (machine learning) and [and] we have a lot of rules and rule-based protections," Neil Kumaran, product manager of Gogole's anti-abuse technology department, told VentureBeat in a telephone interview. "what's interesting about TensorFlow is that it can detect the rest of the spam very well."

This is not to say that Gmail engineers are not using TensorFlow to improve spam detection. Google mentioned the team's migration to the framework in a blog post in May 2017. Instead, according to Kumaran, this latest milestone was achieved through widespread adoption of TensorFlow and a larger set of "large-scale" spam classifications.

"for some time, Gmail has protected users from spam, phishing and malware, and we have used a variety of technologies," he added.

So what kind of incremental spam can make Gmail better? Mainly based on image garbage. Picture spam accounted for 8% of spam traffic in 2011, and text messages were embedded in attached photos. It attempts to bypass text analysis spam filters and optical character recognition (OCR) tools, some of which are quite complex. Liars use obfuscation to prevent OCR tools from reading embedded text, or try to mislead signature-based detection algorithms with meaningless words.

Kumaransaid TensorFlow provided flexibility to the Gmail team in the experiment, allowing faster deployment of new methods to combat image spam and other new types of mail.

"(spammers continue to use) new domain names -. For example, (send) some legitimate emails and then start sending spam.... historically, our employees have been using TensorFlow to get this information," Kumaran said. "it provides us with better tools so that we can experiment more quickly."

Google has long relied on users to determine which emails are real and which are not. It uses this data to train its artificial intelligence system.

The report Spam and non-Spam buttons in Gmail are not only used to display-but also to notify machine learning algorithms for email filtering to improve their ability to detect unwanted messages and reflect personal preferences. (the latter is why some users see weekly newsletters in their inboxes, while others don't. They complement Google's phishing detection model. In some cases, detailed analysis of the message is delayed when its algorithm is updated in real time.

"everyone may have a different definition of spam," says Kumaran. "it can be good for one person or bad for another."it's possible to personalize through machine learning.-- (it) does a very good job of considering users' specific preferences."

Some people may immediately accuse Google of coloring a large number of emails with big brushes, but it occasionally offers an olive branch. In 2015, for example, it released Postmaster Tools, a diagnostic suite that exposes sending errors, spam reports and other potentially useful analyses to companies that send large amounts of email to Gmail users.

Mountain View says it will eventually try to protect users and organizations from cyber attacks. A survey conducted by McAfee in 2015 found that 97% of people could not correctly identify phishing scams. If successful, medium-sized companies lose an average of about $1.6 million in e-mail.

If all goes according to plan, spam detection algorithms will not be the only thing TensorFlow is pushing. Kumaran said the team is already testing improved phishing and malware filtering models and hopes to explain them in detail in the future. "I hope we can publish something similar in the future," Kumaran said.

The blog post. "

Outside the world of message filtering, AI is creeping into all aspects of old and new versions of Gmail. In 2016, Google launched Smart Reply, which uses machine learning to suggest replying to emails, which is being improved and improved as Google introduces it to platforms such as hangouts chat. (according to Google, smart responses process hundreds of millions of messages a day, and more than 10% of e-mail responses are done through mobile phones.) Recently, Google launched Smart Compose, which uses artificial intelligence to automatically complete words and phrases, as well as "nudge", a machine learning-driven feature that provides follow-up email alerts. These are in addition to high-priority notifications, which are settings that limit email notifications to "important messages", while newsletters are unsubscribed recommendations.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.