The COVID-19 pandemic and the presidential election have led to a significant increase in cyberattacks via email, text message and social media aimed at stealing personal information and destroying vital data. To help stop these hackers before they strike, computer science researchers at the University of Houston have been awarded a three-year, $660,000 grant from the U.S. Army Research Office.
Rakesh Verma, computer science professor at the UH College of Natural Sciences and Mathematics and co-principal investigator, said his research team will go beyond commonly-used cyber defense techniques such as honeypots or moving target defense, both focused on fooling the hacker by mimicking likely targets of attacks or increasing uncertainty and complexity.
“Instead, we will generate new attacks of our own. We want to be proactive rather than reactive,” said Verma. “Cyber criminals are getting more creative at each turn, so the idea is to be one step ahead of any type of attack.”
The team of postdoctoral, graduate and undergraduate researchers will design machine learning and natural language processing techniques — a computer system’s ability to read and understand spoken or written language — that can produce unlimited, open-ended attacks. The goal is to generate novel attacks on a daily basis using adversarial machine learning to help develop new, ingenious filters to ward off those attacks.
“We want to close the loop by subjecting our detectors to these new attacks, so the detectors are continuously learning and improving themselves rather than passively waiting for attacks,” said Arjun Mukherjee, principal investigator and associate professor of computer science. “We will compare our techniques against state-of-the-art baselines on diverse datasets in realistic scenarios.”
A recent assessment by the International Criminal Police Organization (INTERPOL) on the impact of COVID-19 on cybercrime showed a significant target shift from individuals and small businesses to major corporations, governments and critical infrastructure. The agency projects a further increase in cybercrime in the future.
“Vulnerabilities related to working from home and the potential for increased financial benefit will see cybercriminals continue to ramp up their activities and develop more advanced and sophisticated modi operandi,” according to the report.
Verma believes at some point online criminals will start to use machine learning and natural language processing to produce cyberattacks.
“We want to have that cycle of automatic improvement, so we create our own attacks and subject our filtering methods to the new attack and see if the new attacks are successful,” Verma explained. “If they are unsuccessful, then we will try to generate better attacks and if they are successful, we will work to improve our filters.”