Ioannidis to Lead $1M NSF Grant for Real-Time Learning for Next Generation Wireless Systems
Current wireless systems need to drastically improve data rates and have ultra-low latency to support pioneering applications such as shared virtual reality experiences and autonomous cars, says Stratis Ioannidis, assistant professor of electrical and computer engineering (ECE) at Northeastern University. He’s the principal investigator on a $1 million National Science Foundation (NSF) grant to meet these goals, in collaboration with ECE Professors Jennifer Dy, Tommaso Melodia, Associate Professor Kaushik Chowdhury, and Assistant Professor Yanzhi Wang.
Next-generation wireless applications crucially rely on the ubiquitous availability and real-time reconfigurability of high-speed wireless links—and this requires wireless devices to perform a broad variety of inference tasks in real-time. “You need systems that are reactive—so they can basically operate under rapidly changing conditions,” says Ioannidis. “For example, part of the channel that’s available right now might stop being available in milliseconds. Or the quality at a certain band may deteriorate because of interference. So you need systems that understand the wireless environment, and can quickly react to changes in it.”
The research team’s proposed solution, “Efficient and Adaptive Real-Time Learning for Next Generation Wireless Systems,” will address the need for faster data and lower latency with a multi-pronged approach that uses the specialties of everyone on the project. “We are a diverse team of Northeastern engineering faculty,” says Ioannidis. “It’s a team of people that work on machine learning, like Jennifer Dy and I; people that work on wireless technologies, like Kaushik Chowdhury and Tommaso Melodia; and Yanzhi Wang who works on speeding up machine learning algorithms using hardware accelerators. Basically, we occupy the entire stack of things you would need in order to perform this research.”
As suggested by Ioannidis, the core of the proposed solution is the machine learning done by wireless systems. It’s not just that the systems need to learn, though—it’s that the algorithm by which they learn needs to adapt to changes just as quickly as the wireless environment itself changes. This is what Ioannidis and other machine learning researchers call “lifelong learning,” or the ability for artificial intelligence to build upon its past knowledge—rather than forgetting what it’s learned before. Lifelong learning adjusts learning algorithms based on this collective knowledge, ultimately optimizing the experience for the end user as learning improves performance and efficiency.
With unprecedented efficiency improvements in next-generation wireless systems so that wireless systems are capable of handling the volume, speed, and unpredictability inherent in the environment, the future of pioneering applications—for consumers and industry—moves closer and faster to reality.
Abstract Source: NSF