Geekin Out: Calculating Breach Likelihood with PGMs

VIDEO DESCRIPTION:

Play, download and edit the free video Geekin Out: Calculating Breach Likelihood with PGMs.

What’s important when creating probabilistic models for breach likelihood and machine learning is explainability. What I mean by explainability is to be able to ask and answer the question “Why is my breach likelihood high?” Just imagine if you had a low credit rating and you’re trying to figure out why it’s low and no one would give you any information.

That would be very frustrating. And similarly, for a security professional if you just tell them that the breach likelihood is high and give them no actionable insight they have no way of making their networks any safer.

Balbix uses a family of machine learning algorithms called probabilistic graphical models and what these models allow you to do is to encode sophisticated domain knowledge and trace how events actually transpire. So you take something like phishing where you have a sequence of events. Being able to tell you exactly which sequence of events led to the other as well as being able to encode the information in a probabilistic manner across these events really enables you to get an accurate score for your likelihood from phishing.

In addition, Balbix uses elements such as vulnerabilities, what global threats are running wild, the exposure of your assets to these threats, the importance of various assets, and any mitigating controls that you have in place to compute the overall breach likelihood.

What’s special about Balbix AI and how we use PGMs is it allows us to do very structured learning. We operate them in two settings. One is an unsupervised setting where this probabilistic graphical model learns what is normal in the network and in other words it builds a mathematical representation of the entire data it observes in the network. And then in the supervised setting it uses the information it has learned to evaluate each and every behavior that it observes in the network and evaluates the breach likelihood across all attack vectors for which we have built PGMs.

What this really helps us to be able to do is separate the normal from the abnormal. And also with the ability to be able to encode security domain knowledge within these models you actually get the best of both worlds. So you can actually get expert security domain knowledge along with advanced mathematics and advanced statistics to be able to really get to a very accurate picture of your breach likelihood.

Download, play and edit free videos and free audios from Geekin Out: Calculating Breach Likelihood with PGMs using RedcoolMedia.net web apps