Mengnan Du's research builds methods and evaluations that make ML systems transparent, fair, robust in vision and language. Focus includes explanations, bias mitigation, shortcut/trojan analysis, misinformation detection, and tools for LLM interpretability.