When he set off for college, Liang Zhao, Emory associate professor and Winship Distinguished Research Professor of computer science, thought he wanted a career in robotics. At Northeastern University in Shengyang, China, he studied automation, then obtained a masters in control engineering.
A highlight of his undergraduate years was competing with classmates in the RoboCup. Students from around the world design small, soccer-playing robots for the annual competition, then face off in tabletop games.
"We all enjoyed playing video games," Zhao says, "so it was like an extension of that."
Many of his classmates focused on building the hardware for the robots. Zhao, however, preferred developing algorithms and software systems to act as the "brain," enabling machines to process sensor data, make decisions and execute actions.
He began to transition into computer science.
"Big data was becoming a hot topic," Zhao recalls. "Humans have accrued so much data we needed to find better ways to process it. The field of AI started moving fast."
The transition wasn't a big leap for Zhao, who says that AI works like robotics: "You give me input and I will give you output to achieve a goal."
AI for science
Zhao went on to get his PhD in computer science at Virginia Polytechnic Institute and State University. He is now a leading expert in AI, data mining and machine learning, pioneering methods to transform research in science, health and industry.
Since joining Emory in 2020, Zhao has built collaborations with scientists across the university, working to speed and improve processes related to discovery and medical diagnostics.
"My interests are broad," he says, "but whenever I can make a contribution to a specialized community, I like to do that."
Most recently, Zhao showed how AI can help speed the design new molecules that boost the power of disinfectants to combat the growing threat of dangerous "superbugs."
For that project, Zhao's lab collaborated with experimental chemists, including Bill Wuest, Emory professor of chemistry, and computational chemists. They created an efficient feedback loop between human expertise and AI, bridging laboratory experimentation and computer engineering.
A range of scientific areas beyond chemistry may benefit from the experimental framework the team developed for the project, Zhao says.
Some examples of Zhao's ongoing projects include the following.
- Working with Michael Treadway, Emory professor of psychology, to boost the throughput of experiments to determine the neurobiology of cost/benefit decision-making in people with diverse mental health profiles. Human experiments are costly and time-consuming, Zhao explains, so the aim is to use data gathered from human participants to build AI agents and simulate experiments.
- Integrating AI into the research of Yang Liu, professor and chair of environmental health at Rollins School of Public Health. AI can help model the dynamics of how plumes of wildfire smoke and other pollutants move through the atmosphere to help understand and protect against the effects on human health.
- Using AI to assist Xiaofeng Yang, professor at Emory Winship Cancer Institute, to support radiology experts in detection of traces of tumors in scanning imagery. "As AI is getting smarter and smarter," says Zhao, "it's becoming increasingly embraced by medical professionals."
- Creating AI tools for analysis of brain health in collaboration with Deqiang Qiu, the program director for magnetic resonance imaging (MRI) at the Emory Center for Systems Imaging Core. Qiu is developing advanced MRI techniques, as well as analytical tools to improve clinical diagnosis and management of neurological diseases, including Alzheimer's disease.
AI is not monolithic, Zhao says. "From a data perspective, it can be understood through several major forms of information: text, spatial data, temporal data and graphs. Together, these data types capture many of the fundamental patterns in the world, as well as humans' conceptual understanding of them.
"For instance," he explains, "we can model a molecule as a network or a graph, similar to how we can model a protein as a network or a graph. Brain functions can also be modeled as a network by graphs."
The first step is to determine the AI subtype appropriate to solve a problem. Next, the problem needs to be expressed in mathematical terms.
"Whatever science domain I work with, I abstract the core problem into mathematical language," Zhao says. "From an AI point of view, the data is just numbers. You see distant science areas become more similar, they are converging towards one another."
AI is poised to deliver enormous benefits to science and society, Zhao says, adding: "I feel fortunate to be living in this age, to witness this change."
