Bingbing Wen

I am a final-year Ph.D. candidate at the University of Washington, where I am advised by Prof. Bill Howe and Prof. Lucy Lu Wang. I am also a member of the UW RAISE Center and collaborate closely with Prof. Yulia Tsvetkov.

My research focuses on the efficiency and reliability of foundation models and agentic systems, aiming to reduce computational overhead while enhancing model trustworthiness. My work is structured around three core pillars:

• Data-Centric Optimization: I develop methods for optimal data mixture selection and curation, designing fine-grained preference signals that align models beyond simple correctness. ICLR2026DATA-FM,COLM25
• Agent Workflows & Modular Architectures: I design adaptive systems that optimize tool use and dynamic routing mechanisms. My work explores how reinforcement learning can orchestrate collaboration among specialized experts to streamline complex agent workflows. TACL2025,ICML2025,IUI2026,CoA
• Reliability-Aware Evaluation: I design frameworks for selective prediction and abstention, enabling models to quantify uncertainty and conserve resources by avoiding unnecessary computation on low-confidence samples. ACL2025, EMNLP2024,Neurips2025