CAREER award recipient to fight rare diseases using AI
College of Engineering Assistant Professor Dawei Zhou wants to make artificial intelligence function more like human intelligence.
While not a perfect system, human reasoning still outshines artificial intelligence (AI) in a number of critical areas. One Virginia Tech researcher wants to help change that.
Insufficient logical reasoning capability in AI can be a disadvantage when trying to tackle complex problems like diagnosing and treating rare diseases and detecting and disrupting financial fraud, said Dawei Zhou, assistant professor in the Department of Computer Science and a core faculty member at the Sanghani Center for Artificial Intelligence and Data Analytics.
Developing AI that can function more like human intelligence and learn from complicated real-world situations is the focus of his recently-announced National Science Foundation Faculty Early Career Development (CAREER) award.
“I am interested in leveraging human-like, conscious thinking to improve the process of machine learning. AI systems like chatGPT are so powerful because they have been trained with vast amounts of data,” Zhou said. “However, the effectiveness of AI is hampered when there is insufficient data to train the models.”
Compared to human intelligence, artificial intelligence learns from the outside world in a very narrow way, Zhou said. Current AI systems can learn complex concepts, but they lack an effective way to generalize knowledge to novel situations and provide reliable and trustworthy services.
Project at a glance
- Title: "Long-Tailed Learning in the Open and Dynamic World: Theories, Algorithms, and Applications"
- Award: CAREER award
- Amount: $600,000
- Timeframe: 2024-29
- Principal investigator: Zhou
The more unusual a phenomenon is, or the more widely it is distributed within a system, the more difficult it is to capture and analyze data about it, Zhou said. The less data we have, the more difficult it is for AI models to learn about the problem and make accurate predictions to help us tackle it.
Current AI methods work well with diseases that occur frequently across populations, such as heart disease and influenza, because there is a trove of data about them, Zhou said. But the models struggle with rare and novel diseases. At the beginning of the COVID-19 pandemic, for example, there was little data to train AI models to help prevent its spread.
It’s the same with financial fraud as health care, and Zhou’s research has similar applications in that arena. The vast majority of financial transactions are legitimate. Fraudulent ones are hard to detect because there are relatively small numbers of them, and they are spread across a sprawling network of institutions and systems.
Zhou has worked extensively with financial institutions in efforts to prevent financial fraud and will tap that network for this project. Rare disease data will come through the Sanghani center’s research relationship with Children’s National Hospital in Washington, D.C.
