Blog Date: 8/11/2016
Author: Ray Coulombe

How your favorite computer apps might surprise you.

In case you missed it, Facebook has artificial intelligence (AI). It’s there to help you tag your friends in photo albums by facial recognition and aids in search queries. And they’re not alone. AI is one of the hottest, fastest growing technologies when it comes to making sense of loads of data.

You, too, can now get deep-learning artificial intelligence systems for your office system through a cloud service. With cloud-based deep learning, companies choose a cloud service and its application programming interfaces for software tasks like recognizing images, filling in search queries, or translating.

In an article from IEEE Spectrum, Zachary Chase Lipton said, “Deep-learning algorithms dominate other machine-learning methods when data sets are large.” The deep-learning researcher works in the Artificial Intelligence Group at the University of California, San Diego, and has examined cloud AI services from companies such as Amazon and IBM. “Thus any company or application that has well-formed prediction problems—such as forecasting demand or translating between languages—could benefit from deep learning.”

But that’s not all; you might even believe computers can start to think because these deep learning programs also have a powerful tool called a “recurrent neural network”—basically the AI “learns” as it mines data. With studies from Lipton, Sharad Vikram, and Julian McAuley at UC San Diego, the research shows these recurrent networks can do everything from process language to reply to email messages.

But how does this neural network work? Basically the networks consist of artificial neurons, which, in biology, are known as the building blocks of the cell. They fire off, or don’t fire off, electrical signals based on signals received from the other neurons attached to them. In your brain, a neuron is connected by a synapse to other neurons. Through these connections, one neuron’s firing can start or stop the firing of other neurons. So, similar to biological neurons, artificial neural network’s stimulated neurons work together. Between each connection in artificial neurons, there is a value assigned called a weight. This weight represents the strength of the connection, where a positive number is a stimulating connection while a negative number is a inhibiting connection. To determine whether this artificial neuron will fire or not, the weight is calculated by the sum of the activations of al the neurons feeding into it. Then that sum is run through an activation function, outputting the desired activation.

Okay now that all the science is behind us, what does this mean for security? For starters, it helps with better analytics. Everything from facial recognition to license plate recognition to non-rules-based algorithms will be analyzed better and faster. Not to mention that there are better analytic measures employed in nearly all aspects of risk analysis.

But it also can better predict events and deployment of security assets by “learning” how the system has reacted to previous events. Finally, there is more effective cyber security overall (due to the AI and other security passes like biometrics) and post-event analysis of the data logs.

Back to Blog List