Yao Zhang
AI Ph.D. Student @ LMU Munich | Stay Hungry, Stay Foolish
Hello there! 👋 I’m Yao, a Ph.D. candidate in Computer Science at LMU Munich under the guidance of Prof. Dr. Volker Tresp. Prior to starting my PhD, I received my master’s degree in 2021 and my bachelor’s degree in 2019, both from LMU Munich, where I specialized in data mining under the guidance of Prof. Dr. Thomas Seidl.
My research focuses on building reasoning-driven and scalable intelligent systems that can perceive, plan, and act autonomously in open environments. I study intelligence as an evolving system that connects adaptive learning, verifiable reasoning, and autonomous system design, aiming to make AI both reliable and deployable in dynamic, multimodal, and real-world settings. These three perspectives form a coherent progression: learning builds the foundation for generalization, reasoning ensures interpretability and trust, and system design integrates these abilities into reliable autonomous behavior.
Research Approach
I approach intelligence research through three interconnected levels, each building upon the others to create robust autonomous systems.
Learning Level
At the learning level, I explore how large models can efficiently adapt to heterogeneous and multimodal data while maintaining robustness, privacy, and scalability. This involves developing lightweight and federated learning frameworks that enable efficient model adaptation across distributed and dynamic environments.
Reasoning Level
At the reasoning level, I study how AI systems can evaluate and align their reasoning processes with decision policies. My work develops process-level modeling and structured feedback mechanisms that connect reasoning reliability with policy optimization, supporting interpretable and verifiable decision-making in autonomous agents.
System Level
At the system level, I design agentic architectures that integrate planning, reflection, and multimodal perception to support long-horizon and collaborative autonomy. This perspective bridges reasoning-grounded decision models with scalable system design, enabling general-purpose web and multimodal agents capable of adaptive and verifiable behavior.
Research Vision
Ultimately, my goal is to bridge research innovation and system architecture, creating AI systems that not only learn and act but also understand, verify, and improve their reasoning, paving the way for trustworthy and scalable autonomous intelligence.
Research Themes
Adaptive and Federated Learning for Scalable Multimodal Intelligence
Efficient, privacy-preserving, and robust adaptation across heterogeneous, multimodal, and dynamic environments.
Process-Level Reasoning and Policy Alignment for Reliable Decision-Making
Modeling reasoning chains and reasoning-conditioned policy learning to achieve interpretable and verifiable autonomy.
Agentic System Architecture and Scalable Multi-Agent Autonomy
Designing planning, reflection, and coordination frameworks that connect reasoning with system-level reliability and scalability.
Feel free to email me if you are interested in my research or just want to chat when visiting Munich ☕. I am always open to new ideas and collaborations.