Trustworthy AIedit
Research topic covering reliability, deletion, fairness, robustness, interpretability, and evaluation.
Trustworthy AI concerns model behavior that can be audited, corrected, updated, or evaluated under realistic constraints.
Research contextedit
Trustworthy AI encompasses machine unlearning, fairness and robustness, interpretability, LLM reliability, and collaborative evaluation. The unifying idea is that reliability is not only a property of a trained model. It also depends on the data process, who can inspect the data, how changes are requested, and how evidence is shared.
Connection to Qiao's workedit
Qiao's work contributes to trustworthy AI through concrete mechanisms. Unlearning papers give methods for deleting or correcting data influence. Synthetic-data work studies how recursive training can fail and how distributed parties can detect the failure. The NeurIPS 2026 paper Illusory Pattern Perception Drives Spurious Inference in Large Language Models examines a different reliability risk: unsupported relational inference from random, unrelated, or insufficient evidence. AI-and-networks projects study how reliability and efficiency change under communication constraints.