The National Science Foundation (NSF) awarded a $600,000 Project Grant under the Computer and Information Science and Engineering (CFDA 47.070) program to the University of South Carolina (USC) to enhance security and mitigate harm in AI-generated vision language models. The 3-year project, running from October 2024 to September 2027, will pursue three technical objectives: 1) developing a prompting framework for detecting harmful content provenance in AI-generated vision language models, 2) implementing machine unlearning techniques to proactively mitigate biases in these models, and 3) detoxifying harmful images through selective blurring. The project aims to address the substantial risks of misuse and potential for harm associated with the rapid growth of AI-generated content, particularly for vulnerable populations. USC will share open-source code, datasets, and live demos as part of this effort to support stakeholders in government, regulatory bodies, and policymaking.
Generated 3/4/25, 4:53 AM