Team
The organizing team was assembled to cover the full failure-mode stack for agentic AI: multi-turn training, evaluation, safety, grounding, security, and community-facing workshop execution.
Northwestern University
Workshop Operations Owner
Zihan Wang is a PhD student at Northwestern University working on reinforcement learning for reasoning agents in multi-turn stochastic environments. He leads overall workshop operations, including submissions, program assembly, and day-of execution.
Canyu Chen is a PhD student at Northwestern University whose research focuses on foundation-model agents, trustworthiness, and multimodality. He leads the contributed-paper workflow, reviewer coordination, and evaluation quality.
David Acuna is a Senior Research Scientist at NVIDIA Research working on reasoning models, synthetic data, inference-time scaling, agents, and reinforcement learning. He supports reviews, speaker coordination, and external outreach.
Jaehun Jung studies how to train and evaluate models with other models under minimal human supervision. He leads evaluation-facing aspects of the workshop, including website artifacts and reproducibility expectations.
Niloofar Mireshghallah works at the intersection of privacy, NLP, and the societal implications of machine learning. She leads discussion design, panel structure, and community-engagement planning for the workshop.
Yejin Choi is a Professor of Computer Science at Stanford and a Senior Fellow at HAI. Her research spans language, reasoning, model behavior, evaluation, and human-centered foundation models and agents.
Dawn Song is a Professor of Computer Science at UC Berkeley whose work focuses on security, privacy, adversarial robustness, and trustworthy machine learning in realistic deployment settings.
Manling Li is an Assistant Professor at Northwestern University whose research focuses on grounded reasoning and multimodal knowledge for foundation-model agents. She leads coverage of world-facing agent failures and grounding-related themes.
The program committee is being assembled with coverage across agents, safety, evaluation, and multimodality.
The current public site reflects the accepted workshop proposal plus organizer-confirmed updates that are ready to share publicly.