9 Ways Artificial Intelligence Can Be Used To Bring IoT, Edge Computing To Life

Making Smart Devices Even More Intelligent

Artificial intelligence (AI) can be injected into every stage of the digital lifecycle, regardless of where the computing is taking place or where the data is being collected.

Some of the IT industry's leading AI experts addressed how artificial intelligence should be incorporated into conversations around the Internet of Things and edge computing during a CRN roundtable discussion last month at XChange Solution Provider 2018 in Orlando, Fla.

Technology leaders from Cybereason, Cylance, IBM, SAS and Aruba, a Hewlett Packard Enterprise Company, address how they're training their AI models for edge computing, the role that data collection and real-time action play around IoT strategy, and why having a robust backend is vital to supporting AI on the edge.

Here's a look at how these vendors are applying artificial intelligence to their IoT and edge computing practices.

Living On The Edge

The Internet of Things uses edge computing to enable smart devices to assist in business workflow management, said David Wilson, vice president of the IBM partner ecosystem. The artificial intelligence component is critical to the overall workload, Wilson said, whether it's out in the sensor, or back in the cloud or data center.

"Wherever the computing is taking place, wherever the data is being collected, we have an opportunity to inject AI into it," Wilson said.

Aruba, a Hewlett Packard Enterprise company, combines artificial intelligence and the Internet of Things through its intelligent edge practice, according to Jisheng Wang, senior director of data science in the CTO office. For instance, solution providers can help put WiFi and computational power into a light bulb or lighting system to make it intelligent, Wang said.

Dealing With Data

Artificial intelligence is only as good as the data you collect, Wang said. Organizations will therefore want to collect network, video and voice data and combine them together to make an intelligent system based in the data center or the cloud, according to Wang.

But 80 percent of data related to IoT devices will be collected, analyzed and then deleted from the edge, Wang said. As a result, organizations cannot afford to send IoT-related data to the cloud as they would other types of data, according to Wang.

"The edge should be intelligent enough to not just collect the data. It should also analyze," he said.

Training The Machines

SAS is very focused on training models, according to Chief Data Scientist Wayne Thompson.

"You build a model offline and then you apply it online for inferencing. And we do that at the edge," he said.

For instance, Thompson said the company has large solar farms at the back of its facility and sells some of the energy back to Duke Power. By training the models, Thompson said SAS is able to look at load factors.

Recurrent networks are currently the best thing for dealing with time-series data, Thompson said.

No Internet Connection Required

Most of Cylance's data processing and model generation happens on the backend, with the company then pushing the data out onto the local IoT devices, according to CTO Rahul Kashyap.

Cylance has a huge farm and cluster on the back end where it trains models, Kashyap said. The company takes the data it has on the back end, Kashyap said, and generates a local model that can be effective either with or without the internet.

"Our solution can be very effective in networks where you have no connectivity to the outside world, or different constrained environments," he said.

Don't Forget The Backend

Like Cylance, a portion of the activity for Cybereason happens on the endpoint and a portion happens on the backend, whether that's on-premise or in the cloud, according to CISO Israel Barak.

Backend processing is vital when attempting to address offensive cyber operations questions related to an attacker that has already gotten into the system, Barak said. The cyber operation can often be detected or prevented in one place, but Barak said organizations are left wondering whether the bad actor has access to other points in their environment.

"It can be detected or prevented in one place, but then you always have to ask yourself the question of, where else are they? Do they have other access points into my environment?" Barak said.

The backend plays an essential role in detecting the potential for lateral movement and creating a broader context around incidents, according to Barak.

Sentiment Analysis

SAS is trying to move its training model into the event stream process engine, Thompson said.

As a conversation is happening in SAS's call center, the company can, in real time, do sentiment analysis to better understand what the customer is doing and help escalate them appropriately, according to Thompson.

"We've ever taken what's called the polarity score from sentiment... and passed that further downstream into a machine-learning model," he said.

Using sentiment analysis as a feature input enable SAS to create better cross-selling models for a client, according to Thompson.

Reacting In Real Time

Data collection and action training is a key piece of Aruba's IoT strategy, Wang said. The other, he said, is around being able to take action in real time.

After training the intelligence model, Wang said organizations want to be able to detect threats inside the enterprise in real-time regardless of whether the data is in the cloud or on the edge. Specifically, companies want to avoid having to send the data to the cloud and then transfer the deletions back to the edge, according to Wang.

"When you try to make those real-time detections of the threats inside the enterprise, you don't want to send the data to the cloud and transfer the deletion back," he said.

The intelligent edge therefore has two roles, Wang said. One is around data collection, Wang said, and the other is focused on the actors in real-time deletions.

Putting Customers In The Driver's Seat

Cylance is very thoughtful around how it collects data and works to give its customers full flexibility around data in hopes of avoiding any potential privacy concerns, Kashyap said.

Customers want to have visibility and controls to allow or disallow what's going out from their environment, according to Kashyap.

"In the world of security, it is very paramount that there are no privacy concerns whatsoever with any customer," he said. "We make sure that we give our customers full flexibility."

Cylance has therefore given customers a lot a control over what data or information is going out, Kashyap said. And if organizations do not want stuff going out, Kashyap said they can completely disallow that as well.

Adding Value Around Alerts

Vendors and partners strive to provide a full and complete attack story in as automatic a fashion as possible when delivering an alert to the end customer, Cybereason's Barak said. This makes it easier for the organization's security operations team to do their job quickly and effectively, according to Barak.

By using automatic processing, Barak said Cybereason can look at other machines and other behaviors and deduce what the context for a particular alert is. This makes it possible for clients to effectively respond to the alert in a single alert, Barak said, and avoid having to play whack-a-mole with the attacker.

"When you provide an alert, you [want to] tell them as automatically as possible... what the attack story is," he said.