With the rapid advancements in AI, creating hyper-realistic images and convincing text has become easier than ever, bringing both innovation and challenges. This capability, however, raises significant concerns, especially in the context of "fake news" or disinformation campaigns, where AI-generated content can appear indistinguishably real.
We are thrilled to have Runsheng Huang, author of the pioneering research paper, "MiRAGeNews: Multimodal Realistic AI-Generated News Detection." Runsheng's research introduces the MiRAGeNews dataset, a first-of-its-kind collection designed specifically to tackle AI-generated disinformation. This dataset contains over 12,500 real and AI-generated image-caption pairs from cutting-edge generators like Midjourney, which are notoriously difficult for both humans and state-of-the-art models to differentiate. By training the MiRAGe detector, which combines image and text analysis, his team has achieved detection performance improvements, including robust handling of out-of-domain data from unseen news sources and AI models.
In this session, Runsheng will delve into the technical and practical aspects of this research, sharing insights into MiRAGe's methodology, its dataset, and the model's capacity to navigate and counteract realistic, multimodal AI-generated content.
Runsheng (Anson) Huang earned his Master’s in Data Science at the University of Pennsylvania and is currently a research assistant advised by Dr. Chris Callison-Burch. He is broadly interested in building socially responsible and safe AI. His current research focuses on robust detection methods for harmful AI-generated content, including realistic fake images, multimodal fake news, and AI-edited images. His recent work MiRAGeNews has been accepted to EMNLP 2024.
©2024 SSI Club. All Rights Reserved