PhD Student | National University of Singapore
Multimodal AI · Trustworthy ML · Misinformation Detection
I am a PhD student in the School of Computing at the National University of Singapore, supervised by Prof. Mohan Kankanhalli. My research focuses on multimodal misinformation, explainable fact-checking, and trustworthy large multimodal models.
My current work studies how to build reliable benchmarks and evidence-grounded systems for detecting and explaining multimodal misinformation in real-world settings.
Prior to joining NUS, I received both my B.Sc. and M.Sc. degrees in Computer Science from Wuhan University, China.
Released the RW-Post dataset and codebase on GitHub: AgentFact.
Our paper Modeling Human Responses to Multimodal AI Content was selected as a Cover Feature by IEEE Computer.
RW-Post was accepted to the CVPR 2026 Workshop on Misinformation Detection in Society PP-MisDet.
A New Dataset and Benchmark for Grounding Multimodal Misinformation was accepted to ACM Multimedia 2025.
3MFact was accepted to AAAI 2025. Co-first author
Combating Misinformation in the Era of Generative AI Models was accepted to ACM Multimedia 2023.
We present RW-Post, a real-world multimodal fact-checking benchmark that connects social media posts with auditable reasoning traces and evidence grounded in human fact-check articles. Using AgentFact, we evaluate modern LVLMs and show that faithful evidence grounding remains a major challenge.
Rather than focusing solely on detecting AI-generated content, this work studies how people perceive, trust, and interact with multimodal AI content. We introduce MhAIM, a large-scale dataset of human- and AI-generated media, and propose T-Lens, a human-centric framework for predicting user responses and providing interpretable explanations of trust, credibility, and influence.
We present GroundLie360, the first real-world benchmark for grounding multimodal misinformation in videos. Beyond detection, this work localizes misinformation across text, speech, and visual modalities with evidence-based explanations, and introduces FakeMark as a VLM-based baseline.
We introduce a video fact-checking dataset with veracity labels, supporting evidence, and human-annotated rationales. We further propose 3MFact, a multi-role multimodal framework that retrieves and synthesizes online evidence to generate predictions, explanations, and supporting evidence.
We examine the challenges introduced by AI-generated misinformation and analyze manipulation traces across multiple levels, from signals and perception to semantics and human cognition. This work provides a conceptual foundation for explainable multimodal misinformation detection in the era of generative AI.
Feel free to reach out for research collaborations, academic discussions, or opportunities related to multimodal AI and trustworthy AI.
I am interested in postdoctoral and research scientist opportunities starting in late 2026 or 2027.