Kaiyan Zhang (张开颜)

I am a final-year PhD candidate at the Department of Electronic Engineering, Tsinghua University, under the guidance of Professor Bowen Zhou. I earned B.S. (2020) and M.S. (2022) degrees in Computer Science and Technology from the Harbin Institute of Technology (HIT), where I was supervised by Weinan Zhang and Ting Liu in the HIT-SCIR lab.

My current passion is pushing the boundaries of domain-specific superintelligence (called ExpertAGI), enabling AI systems to achieve expert-level reasoning and collaboration across high-value and practical scenarios. My research directions include:

• Scalable Learning (e.g., RL): Developing novel frameworks for scalable reinforcement learning, such as TTRL (test-time RL with unlabeled data), SSRL (self-search RL leveraging intrinsic model capabilities), MARTI (multi-agent RL coordination), and OpenPRM (scalable process reward modeling), all aiming to reduce supervision costs and unlock self-improving LLMs.

• Collaborative Intelligence: Designing mechanisms for model cooperation and synergy, including CRaSh (efficient fine-tuning via clustering and sharing), CoGenesis (secure collaboration between large and small models), FS-Gen (unified laws in collaborative decoding), and MARTI, to empower collective intelligence among agents.

• Scientific Intelligence: Applying LLMs to scientific discovery, with projects like UltraMedical (generalist biomedical models), hypothesis proposer (autonomous scientific hypothesis generation), and ReviewRL (reinforcement learning for automated scientific review), advancing AI’s role in research and innovation.

Expect to graduate in June 2026. My CV is here.