Mellen Y. Pu

   “Through rigorous science, the world aligns; I cherish unveiling its designs. “

—— Mellen Y. Pu

About

“How can AI discover new scientific mechanisms?” This is the fundamental question driving his research.

In 2025, he introduced scientific principles as optimizable objects within multi-agent systems for scientific discovery, treating them as the medium for discovery. This led to a novel method that iteratively updates hypotheses by accumulating experimental evidence to optimize underlying principles (Pu et al., 2025). He subsequently expanded this into a unified framework based on dynamic principle space evolution (Pu et al., 2026). This system not only leverages existing scientific laws for exploration but also employs information-theoretic approaches to update the principle space itself, enabling anomaly-driven principle augmentation.

Mellen Y. Pu is a Ph.D. candidate in the School of Engineering at Westlake University, under the joint mentorship of Prof. Tao Lin and Fellow of the Royal Society of Chemistry, Prof. Hongyu Chen. He is currently focusing on Principle-Aware Scientific Discovery. He looks forward to long-term cooperation and extensive exchanges on Agentic Learning and Evolving in general domains.