Jeremy McMahan

Hi! I am a Computer Science PhD graduate from the University of Wisconsin-Madison, where I was advised by Jerry Zhu and worked closely with Qiaomin Xie and Yudong Chen.

I develop algorithmic foundations for capable and reliable AI agents - agents that can solve complex decision-making tasks while operating safely under uncertainty and adversarial interactions. My work leverages machine learning to handle uncertainty and worst-case algorithm design to ensure performance and safety. This yields the first polynomial-time algorithms for general constrained and robust multi-agent reinforcement learning — overcoming intractability barriers to enable near-optimal agents that satisfy safety constraints and operate robustly, even in NP-hard domains.

More broadly, these results show that reliability and capability can be tractably achieved simultaneously, and I'm excited to extend these principles to more general agent architectures and hybrid AI decision-making systems.