Apr 05 2021

Researchers Eye Machine Learning to Secure IoT Data

A federated learning model offers new possibilities for keeping information collected from the Internet of Things anonymous and private.

Traffic cameras that sit on the edge of the Internet of Things may contain sensitive data, as may smart home energy meters. Even smart parking meters may reveal the locations of vehicles, identified by their license plates, and that information should be secured.

Researchers at the Missouri University of Science and Technology are examining how machine learning might secure IoT data. The National Science Foundation is funding work to create new federated learning algorithms. Sajal Das, the Daniel C. St. Clair chair of computer science at Missouri S&T, and Tony Luo, an associate professor of computer science, lead the effort. 

“Sensors and IoT are playing a significant role in the data collection process,” Das says. Vulnerabilities can result from IoT devices transmitting through a “data collection point” in the cloud. ML and analytic algorithms process data collected at the edge through IoT devices, and adversaries can compromise these algorithms. 

The Internet of Things Captures Data Spread Across Sources

Just as a federated government system consists of a central government and smaller political entities, a federated learning algorithm collects data from several edge devices rather than from a single database. Federated learning involves training data and storing it locally. By eliminating the sharing of IoT data, it becomes secure. 

“Federated learning means you are learning by distributing the data to different stakeholders without revealing the identity and the sources of the data,” Das says.

He adds, “The algorithm itself gives the power because of the way we develop the algorithm. We are not sharing the entire data set and the content to every device or every stakeholder, and everybody only sees a partial view of it.” 

Federated learning holds potential for protecting data for air quality monitoring, energy management and overall daily living in a city, Das says. ML algorithms could also secure data in transportation systems, protect critical infrastructure and serve police departments and emergency personnel.

MORE FROM STATETECH: How can you best protect operational technology in smart cities? 

Smart Cities Could Augment Cybersecurity of Federated Data

Aaron Deacon, managing director of Kansas City Digital Drive, a Missouri nonprofit at the intersection of emerging tech and civic good, sees the potential value of decentralization in federated learning as a strategy for securing IoT data. He compares federated learning to a black box concept in which a sensor is monitored without its internal data being viewed. 

“That’s what happens at the edge of the network,” Deacon says. “The question is, how do you preserve that and honor that value with the competing value of everything being centralized and being able to work together — which is where you gain the efficiencies, but it opens you up to vulnerabilities.” 

Smart cities can also use ML as part of an intrusion detection strategy to keep an eye on data traffic and react to security issues, according to David Evans, CIO of Kansas City, Mo

“Device counts are growing on the internet, and IoT is going to continue to grow drastically,” Evans says. “I think the majority of that is going to be in the private home use area and cybersecurity data collection and monitoring.” 

Another key area for growth of ML in smart cities may be using license plate readers to help solve crimes. A potential area for development is consolidating logging and alert services for the government and utility companies. Although ML has real potential for securing smart cities, the technology is still in development mode, Evans says.

tioloco/Getty Images