Explainable AI Team (Talk by Qingcheng Zeng, Northwestern University)

The Explainable AI Team at RIKEN AIP hosts a one-hour research talk by Qingcheng Zeng of Northwestern University, titled "Reliable Embedding Systems for Real-World Search and Discovery". The session runs online from 10:00 to 11:00 JST on Thursday, September 3, 2026. Text embeddings now sit between language and computation, powering search and large-scale text analysis, but real information needs rarely match the clean uniform shape assumed by standard benchmarks. People move across languages, express nuanced preferences, and read meaning through domain and social context. Zeng argues that reliable search and discovery needs more than larger models and generic semantic similarity. Drawing on his own work, the talk shows how mixed-language queries expose hidden representational failures, how targeted training makes retrieval respond to user-defined criteria, and how context-sensitive representations turn embeddings into interpretable instruments for studying social meaning. The through-line is adaptive embedding systems that account for language, intent, and domain knowledge as information needs change.