Haoquan Zhang 张皓泉 CUHK · ByteDance Seed ByteDance Seed CUHK

My research interests are Agentic AI and visual generation. In Agentic AI, I explore environment and reward design to equip agents with more general capabilities. I also do benchmarking to define the capability boundaries of agentic AI. You can check our work at Agent-X. In visual generation, I focus on understanding and improving the generation process for better visual synthesis.

I'm also intrigued by the representations learned by models. I believe it is crucial to carefully observe and analyze these internal structures to better understand the inner mechanisms of neural networks.