A Sage approach to edge computing
While smart sensors provide unprecedented opportunities to better understand the world around us — from storms to wildfires, traffic to toxins — they present a host of challenges in how best to analyze and use the immense volume of high-quality streaming data they collect.
To overcome these challenges, the National Science Foundation awarded a $9.3 million grant to launch Sage, a novel project to build a continent-spanning network of smart sensors and exploit dramatic improvements in artificial intelligence technology.
Sage is led by a team of experts from the Northwestern-Argonne Institute of Science and Engineering (NAISE), a collaboration between Northwestern University and the U.S. Department of Energy's (DOE) Argonne National Laboratory.
Previous solutions devised to address the data deluge from distributed sensor systems — periodic retrieval from a hard drive for analysis or capturing and using only a small portion of the data — have not taken full advantage of the sensors' real-time data-collection capabilities. The Sage project will embed small computers directly into the sensor network, relying on advances in artificial intelligence and edge computing to analyze the sensor data in real time.
The project is led by NAISE's Pete Beckman and includes a large cast of Argonne players. The underlying technology behind Sage — the Array of Things and the Waggle nodes — was developed at Argonne by Beckman and colleagues Charlie Catlett and Raj Sankaran.
Scott Collis, of Argonne's Environmental Science Division, serves as a science consultant on the project, responsible for determining how the cyberinfrastructure and supporting technologies developed during the project can best serve the atmospheric science community.
The individual edge computers developed for Sage will act as hosts for artificial intelligence, especially computer vision. Collis and his team are currently working to ensure that Sage is using the correct camera technology — technology that can be used to answer multiple science questions.
“We have these pan-tilt-zoom cameras that allow users, for instance, to zoom out and identify rain on the ground or traffic patterns and then zoom in to track a particular cloud as it moves across the sky,” Collis said. “We want to build a database — a cloud atlas — of individual cloud elements. To do that, we need a machine learning algorithm to identify the clouds and predict their motion, telling the camera center where to point, how far to zoom in, where to set the exposure and so on.”
Part of Argonne's role, therefore — beyond having developed the technology for Sage — is devising the algorithms for the camera centers.
Collis and his colleagues are also involved in a sister project to Sage called ARMing the Edge. The project is an effort by DOE to use the cyberinfrastructure developed for Sage for the benefit of its Atmospheric Radiation Measurement (ARM) program. Conducted in parallel with Sage, the project is a great example of leveraging the high-risk, high-reward NSF and DOE research to enhance science in the national interest.