As attackers become more and more sophisticated, it is necessary to have advanced technologies that can detect attacks that traditional tools can not.

This is the mission of Blue Hexagon, which came out furtively on February 5th with a deep learning platform to help automatically detect potential threats. The Blue Hexagon platform offers the promise of threat prevention in near real-time, with visibility into compromising indicators for an attack.

"Our technology is rather unique in that we have adopted a technology that is very effective for computer vision and speech and that we are applying it to the complex problem of computer security," said Nayeem Islam, co-founder. and CEO of Blue Hexagon. told eWEEK.

Over the last ten years, Mr. Islam had worked at Qualcomm, where he led a research and development group involved in deep learning and its implementation on mobile phones. He left Qualcomm to create Blue Hexagon in June 2017. Along with its debut, Blue Hexagon also announced a $ 31 million fundraiser from Benchmark and Altimeter to help develop technology and commercialization efforts.

What is in-depth learning?

The terms artificial intelligence (AI), machine learning (ML) and deep learning are often used as synonyms, but there is a fundamental difference between what different technologies allow and how they work.

Islam explained that with classical machine learning, it is necessary to perform what is called engineering features, which means that you need to know a lot about the problem and teach the algorithm to look for elements, in a process that often involves a lot of engineering and human intervention.

"In-depth learning is a huge advance in that the engineer does not have to provide a lot of information on what to look for." The algorithm find things by itself, "explained Islam. "This is one of the most important aspects of deep learning and why it is so effective at solving a problem such as network security, which is very complicated with many forms of malicious programs that circulate on the network. "

With machine learning, there is often a distinction between supervised and unsupervised learning, when a system is directed to a set of data to be learned. Islam said that the deep learning model used at Blue Hexagon is a supervised technique. He explained that the real problem was how much work the engineer has to do to tell the system what to look for, in terms of threat and non-threat.

"So we have data that we use to train our algorithms, and we generate what we call inference models," Islam said.

He added that with other techniques, a developer should instruct the algorithm to look for very specific items to determine what a threat is. With Blue Hexagon's in-depth learning model, Islam said that once the algorithm understood that something is a virus or something that is not a virus, it's not a virus. 39 is everything he needs to know to work.

"We are really focused on network threats that are very specifically targeted, like malware variants," said Islam. "But deep learning can apply to many types of threats."

Sean Michael Kerner is Editor-in-Chief at eWEEK and InternetNews.com. Follow him on Twitter @TechJournalist.