Assistant professor Roberto Yus aims to bridge the gap between the raw data streaming from billions of sensors and the answers to questions people are actually asking.
Billions of internet-connected “smart” devices and sensors—such as video doorbells, automated thermostats, and motion sensors—are deployed in the world today, and the number keeps growing. But most of the data these devices produce never actually gets used in a meaningful way, says Roberto Yus, an assistant professor in computer science and electrical engineering.
Right now, most companies or organizations need teams of highly trained data scientists to derive actionable intelligence from their sensors, Yus says. “And when they don’t have that, all that data just sits there unused.”
Yus would like to deliver the promised benefits of a wired up world to a larger segment of humanity. His research touches key areas needed to address this challenge, such as how to provide computers with the structures necessary to store, organize, and reason about complex data, and how to protect privacy and security. He recently received a prestigious National Science Foundation CAREER award to further this goal.
Q: What is the main challenge holding back more widespread benefits from the Internet of Things?
A: There is a gap between the raw data that these sensors and devices produce and the way humans communicate.
The data are very raw—images, audio, sometimes just strings of numbers. And the information people want is much higher level. For example, a school principal may want to know whether classrooms have good enough ventilation during the hours when students are actually in them. Or you might imagine a nurse in a hospital who wants to know where the wheelchairs are. Or a homeowner who wants to know whether their house is actually energy-efficient, or where in the home they are wasting the most.
My research combines AI and data management techniques to bridge this gap. The system sits in the layer between the sensors and the people, and figures out automatically which sensors to look at, which AI models to apply, and how to combine the results to answer the questions people have. I call this the “sensorless” Internet of Things. It still has sensors, but people can get what they need without ever having to think about a sensor or a smart device.
Right now, AI assistants can talk about the world but they cannot really perceive it that well. The sensorless IoT is the infrastructure that will let AI understand what is actually happening around us, in real time and with the proper privacy protections.
Q: What are some of the obstacles the research will have to tackle?
A: First off, the sensors and devices deployed in a smart space are very different and produce lots of different types of data. This makes it very challenging to have general solutions that work for all smart spaces.
The other big complexity is navigating the trade-offs between accuracy, speed, cost, and privacy. And sometimes these trade-offs need to change. For example, if there’s a fire, maybe the privacy concerns go down in priority because you want to get the most accurate count of people in every room, and you want to get it fast.
Q: In general, the privacy concern seems like a big one. How do you handle it?
A: The privacy part of this is very important. If you start deploying sensors without telling people what is going on, there will be pushback.
For our own projects, we are thinking already about how to make sure that things are transparent. Maybe whenever we deploy a sensor, we also put a QR code that people can scan to find out more information. And we are thinking about how organizations like the university can create rules about how the data can be stored and used and how the system can automatically enforce those rules—so compliance is built in instead of just promised.
You have to have people buying in, otherwise the project will not be successful.
Q: What critical thinking role will humans continue to play in this picture?
A: A human has to stay in the loop and be informed so they can make judgement calls. So, for example, a system might decide that the best way to know how many people are in a building is to tap into all of the cameras. But the human needs to stay in the loop and might decide, “No, I don’t want you doing that.” And then the human gives directions to count how many people are in a room without identifying who they are.
Q: Part of the NSF CAREER award is dedicated to outreach and education efforts. How do you plan to involve undergraduate students?
A: Since I joined UMBC, I’ve been involving undergrads in research, and it’s been great. One of the things that I always challenge them to do is figure out the pain points of the campus, and come up with solutions.
So, for example, students have told me that sometimes when they go to Chick-fil-A or Starbucks, the queue is massive and they don’t have time to wait— wouldn’t it be great if they could check the queue before they go?
I’m developing a new course focused on designing and deploying smart services for the campus. I want the students to come up with their own projects. And I’m very excited to see their ideas!
Beyond UMBC, the award will also support hands-on IoT projects at two Maryland high schools and a summer camp for Baltimore-area students. The idea is to get young people building real systems that improve their own schools and communities, so they can see themselves as future scientists and engineers.
