Computer Scientist Works to Strengthen Artificial Intelligence Models
Associate Professor Ming Shao from the Miner School received a nearly $500,000 grant from the NSF to make multimodal AI systems safer against data contamination and manipulation.
Ming Shao, an associate professor at the Miner School of Computer and Information Sciences, has received a nearly $500,000 National Science Foundation CAREER grant to make artificial intelligence models that process multiple data formats, such as text, imagery, audio and video, safer.
Delusions in Artificial Intelligence Models
Generative AI tools sometimes misinterpret images and produce erroneous results. Minor changes or deliberate manipulations can cause systems to completely misidentify objects.
Details of the NSF CAREER Grant
Associate Professor Ming Shao has been awarded a prestigious CAREER grant of approximately $500,000 from the National Science Foundation to increase the reliability of artificial intelligence tools.
Contributing to the Educational Program
Aimed at transferring his research findings and knowledge to students, Shao also leads the administration of the newly launched Applied Artificial Intelligence and Data Science program within the Miner School.
Data Contamination and Cyber Threats
Stating that data is the greatest power behind AI models, experts point out that dirty data, such as outdated texts or low-quality photos, can cause flawed outputs.
Visual and Language Models Research
Shao published the results of his research examining attacks on systems processing visual and text data in the scientific journal "Neural Networks" and analyzed the types of attacks.
Resilience Against Attacks
Researchers aim to make artificial intelligence more resistant to these threats by training it with potential attack types it may encounter, and to ensure that different data formats support one another.
Continuous Learning and Memory
Aiming to develop a continuously learning model so that systems can stay up to date, Shao seeks to minimize the problem of forgetting old information while learning new data.
Studies Conducted with Students
While undergraduate and graduate students are included in the research team, two Ph.D. students are also actively working on the project within the scope of the NSF grant.