Cornell Bowers College of Computing and Information Science
A color graphic showing photos of two men and the DARPA logo

Story

Computer scientists awarded $3M to bolster cybersecurity

October 4, 2023

By Louis DiPietro

A team of Cornell computer scientists has been awarded a $3 million grant from the Defense Advanced Research Projects Agency (DARPA), the research and development arm of the U.S. Department of Defense, to leverage reinforcement learning to make computer networks stronger, dynamic and more secure.

The project is called LANCER (LeArning Network CybERagents), and researchers hope it will yield smarter, dynamic defenses for cybersecurity experts in the perpetual cat-and-mouse game between defenders and attackers.

Armed with reams of network data and using a process called reinforcement learning, the Cornell team will essentially create artificial intelligence versions of the cat and the mouse, and then pit them against each other. The idea is to allow these AI models to train each other, ultimately enabling defenders to predict attack sequences and sniff out attacks faster and without human intervention. This AI-powered defender model offers an automated and proactive tool to potentially shore up computer networks everywhere.

“The key benefit of reinforcement learning is that it’s potentially capable of learning defense strategies that are more sophisticated than the ones used by humans today,” said Wen Sun, assistant professor of computer science in the Cornell Ann S. Bowers College of Computing and Information Science and project co-investigator.

Similar to how ChatGPT, another AI model, now offers impressive performance on many language-related tasks, “we are hoping that we can see such a revolution in network defense as well,” he said.

“LANCER builds on several decades of work by the community seeking to make networks fully programmable, top to bottom and end to end,” said Nate Foster, professor of computer science and the project’s principal investigator. “Now the question is, how can we use programmability to help secure networks? Reinforcement learning offers powerful tools, but the challenge, as with any AI-based approach, lies in developing mechanisms that are provably robust, even in adversarial settings like network security."

Once completed, Foster and Sun intend to release all LANCER software and datasets as open-source software.

Launched earlier this month, LANCER is the latest DARPA project for Foster, a field leader in programming languages and software-defined networks. In 2020, he was part of a research team that received a $30 million DARPA grant to develop Pronto, a fully programmable computer network. LANCER will build on Pronto’s software and hardware infrastructure to help develop and train defensive agents using reinforcement learning.

Louis DiPietro is a writer in the Cornell Ann S. Bowers College of Computing and Information Science.

This story was originally published in the Cornell Chronicle.