News

Toward Trustworthy Machine Learning Models and More Actionable Decision-Making

Large Language Models (LLMs) and reinforcement learning (RL) hold great promise for decision-making, but concerns about trustworthiness and real-world transfer remain. This talk introduces Topology-based Uncertainty Quantification (Topo-UQ), which models reasoning as topologies to reveal inconsistencies beyond answer-level uncertainty, and Prompt-to-Transfer (PromptGAT), which uses LLM prompts to bridge the Sim-to-Real gap in RL solutions for traffic signal control. Together, these works highlight a path from assessing when to trust AI systems to enabling their actionability in real-world settings.

Longchao is a fourth-year Ph.D. candidate at Arizona State University. His research interests are data mining, RL, and trustworthy AI. He has been working on the trustworthiness of the Machine Learning Models, and actionable decision makings, with publications in top venues such as NeurIPS, ICML, AAAI, KDD, CIKM, ECML-PKDD, Machine Learning, and SDM, etc. He has obtained 2025 Google Fellowship nomination, twice school fellowship award and SDM'25 best poster award.