Vasisht, Singh Connect Next Generation Networks Technology to Industry Collaborators Through NSF RINGS Program

8/24/2022 8:58:02 AM Aaron Seidlitz, Illinois CS

Illinois CS professors Deepak Vasisht (left) in a blue suit and Gagandeep Singh in a floral print shirt.
Illinois CS professors Deepak Vasisht (left) and Gagandeep Singh brought their expertise in wireless networking as well as machine learning, respectively, to this project.

Illinois Computer Science professors Deepak Vasisht and Gagandeep Singh, partnering with Illinois CS affiliate and Electrical & Computer Engineering faculty member Haitham Hassanieh, are about to combine their expertise in wireless systems and machine learning (ML) to further delve into what’s possible through Next Generation (NextG) wireless and mobile communication.

Their goal is to help deliver new possibilities for end users in this realm of CS research, ranging from connectivity for virtual reality headsets or the metaverse to reliable connectivity for autonomous cars.

In July, they might have found the exact right outlet for this opportunity through a new NSF grant program called Resilient & Intelligent NextG Systems (RINGS). Over the next three years, Vasisht, Singh, and Hassanieh will work on a grant that comes with nearly $800,000 in funding.

Haitham Hassanieh
Haitham Hassanieh

Their project is entitled “Provably Robust Machine Learning for Next Generation Cellular Networks,” and focuses on state-of-the-art Machine Learning frameworks that will be key enablers for Next-generation (NextG) networks.”

What makes the RINGS program unique is that it directly includes nine industry collaborators – including Apple, Ericsson, VMware, Microsoft, Qualcomm, Google, IBM, and more. – with helpful oversight from the NSF and the National Institute of Standards and Technology.

“The exciting idea is that basically the NSF and industry came together to create this program built on a three-way collaboration between the government, industry, and academia,” said Vasisht, who is the project’s primary investigator. “The first piece to our work focuses on figuring out where machine learning works and where it doesn’t. Also, for other wireless companies like – Qualcomm and Apple and Ericsson – they want to figure out how to use Machine Learning and AI to build the next generation of wireless networks.”

Vasisht explained that there are three core elements to their project that will showcase the way their efforts can benefit the next generation of wireless systems.

The first two elements focus on building the core technological capabilities. This includes communications – or the way data travels from point A to point B – and sensing – which is about using radio signals to sense the environment around them.

“Networks of the future will have many new applications like virtual reality, autonomous driving, large scale sensor deployments. You also have to consider that the networks will have increasing complexity due to hardware – more antennas, more spectrum bands, etc. What we are trying to say is that to meet the needs, by exploiting these hardware capabilities, you need something in the middle that manages the complex hardware and delivers the new applications,” Vasisht said. “To us, that thing is machine learning.”

The third core element addresses a problem of modern ML methods: a lack of trust in their behavior on unseen inputs.

“For trustworthy adoption of ML-enabled NextG systems, it is essential that the deployed models come with formal guarantees about their robustness on an infinite set of unseen inputs. The traditional methods cannot ensure this; therefore, we need to create new technology to obtain trustworthy ML models,” said Singh.

The work in ML robustness might take more time to develop than the prongs of the project focused on communications and sensing.

However, Singh is thrilled about the potential impact of machine learning with formal robustness guarantees  in this area. His recent collaboration with Vasisht and PhD student Zikun Li presented at a leading wireless conference introduced FIRE as an effort to solve critical bottlenecks for the advancement of next generation wireless systems through the use of machine learning

“The FIRE project started exciting work for me, as it first showed we could utilize machine learning in an end-to-end setting like wireless systems,” Singh said. “Once we figured out that was possible, the RINGS project will help take the next step by connecting us with potential industry partners to ensure an outlet for this technology.”

In preparation for submitting their proposal for the RINGS grant, Singh and Vasisht also met with Hassanieh – whose research focuses on wireless networking, IoT and mobile systems, wireless imaging and sensing, and sparse recovery algorithms and applications.

“What we decided after meeting together was simple,” Vasisht said. “We just thought, ‘Let’s go for it.’”