I am a Ph.D. candidate at X-Lance Lab, Shanghai Jiao Tong University, under supervision of Prof. Kai Yu, majoring in computer science and technology. Before that, I received my B.S. degree from the Department of Automation, Tsinghua University in 2020. In the first four months of 2024, I worked as a research assisstant with Prof. Tao Yu at XLANG Lab, the University of Hong Kong.

My research interest focuses on text-rich visual UI interaction. Currently, Iā€™m working on constrution of realistic, complex interaction benchmark for GUI interaction. Iā€™m also studying how to design smarter GUI agents with reinforcement learning (RL) and large language models (LLM), or by combining both.

šŸ“ Publications

NeurIPS 2023
Rememberer [NeurIPS 2023]

Large Language Models Are Semi-Parametric Reinforcement Learning Agents

Danyang Zhang, Lu Chen, Situo Zhang, Hongshen Xu, Zihan Zhao, Kai Yu
NeurIPS 2023 | Code

  • We designed Rememberer, a novel evolvable LLM-based agent framework, by equipping the LLM with a long-term experience memory, so as to enable the LLM to exploit the interaction experiences to improve performance.
  • We proposed Reinforcement Learning with Experience Memory (RLEM) to update the memory, so that the agent can learn from both success and failure, and evolve its capability without fine-tuning LLM parameters.
  • Rememberer demonstrates superior performance and robustness on both WebShop and WikiHow task set.
Mobile-Env

Mobile-Env: An Evaluation Platform and Benchmark for Interactive Agents in LLM Era

Danyang Zhang, Lu Chen, Zihan Zhao, Ruisheng Cao, Kai Yu
Project | Task Set

We designed an easily-extensible, adaptable, and close-to-reality interaction platform based on Android Mobile.

OSWorld

OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments

Tianbao Xie, Danyang Zhang, Jixuan Chen, Xiaochuan Li, Siheng Zhao, Ruisheng Cao, Toh Jing Hua, Zhoujun Cheng, Dongchan Shin, Fangyu Lei, Yitao Liu, Yiheng Xu, Shuyan Zhou, Silvio Savarese, Caiming Xiong, Victor Zhong, Tao Yu
Project | Code

We designed a unified interaction benchmark for real-world desktop tasks.

šŸ“– Educations

  • 2020.9-(2025.6), Ph.D., SEIEE, Shanghai Jiao Tong University
  • 2016.8-2020.6, B.S., Department of Automation, Tsinghua University

šŸŽ– Honors and Awards

  • 2020.9-2025.6, The 2nd Wu Wenjun AI Honorary Doctoral Progam
  • 2020.9-2025.6, Zhang Xu Scholarship
  • 2018.10, Academic Excellence Scholarship 2017~2018
  • 2017.10, Academic Excellence Scholarship 2017~2018