Xinyuan Wang

I am an Ph.D. student at HKU, mentored by Prof. Tao Yu. I obtained my master’s degree from University of California, San Diego (UCSD). I was luckily to be mentored two distinguished professors at UCSD in Natural Language Processing and Computer Vision - Prof. Zhiting Hu and Prof. Zhuowen Tu. Prior to my study at UCSD, I graduated from Central South University (CSU) in Hunan, China, where I was mentored by Prof. Ying Zhao.

Research Interests

Agent Foundation Model: Designing and developing LLM/VLM based agent foundation model capable of interpreting and executing actions across real-world, digital, and simulated environments.
Language Model Reasoning: Improving the planning, reasoning, decision-making capability of VLM/LLMs . (LLM Reasoners)
Foundation Model Prompting: Employing interpretable prompting to bridge the domain gap between user objectives and the outputs of foundation models. Effectively boosting the performance of foundation models on complex tasks through efficient and effective prompting. (PromptAgent)

Research Overview

I am now working on computer-use agent foundation model, supervised by Prof. Tao Yu. In Prof. Zhiting Hu’s group, I worked on automatic LLM prompt optimization with Zhen. Recently our paper PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization is accepted by ICLR 2024. I am also working on LLM Reasoning by contributing to the LLM Reasoners library, which ensembles the most recent LLM reasoning methods and models. In Prof. Zhuowen Tu’s group, we are working on how to inprove diffusion models’ conceptual performance with an end-to-end loss. During my undergraduate years, I was mentored by Prof. Ying Zhao and worked on Interpretation of Convolutional Neural Networks and Visualization. Here is my graduate thesis: The Research on The Interpretability Method of DeepNeural Network Based on Average Image

How to contact me

Email: xywang626@gmail.com

News

Apr 15, 2025	Kimi-VL Technical Report is publish on Arxiv! I worked on its computer-use capability as a core contributor.
Apr 8, 2024	LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Models is accepted by ICLR2024 workshop!
Jan 16, 2024	PromptAgent is accepted by ICLR 2024 (The Twelfth International Conference on Learning Representations)!
Nov 17, 2023	PromptAgent’s poster is presented at SoCal NLP 2023 at UCLA, Los Angeles, CA!
Oct 25, 2023	Paper published on Arxiv! PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization

Selected Publications

Kimi-vl technical report

Kimi Team, Angang Du, Bohong Yin, and 8 more authors

arXiv preprint arXiv:2504.07491, 2025
LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Models

Shibo Hao, Yi Gu, Haotian Luo, and 8 more authors

arXiv preprint arXiv:2404.05221, 2024

Abs PDF Code

Generating accurate step-by-step reasoning is essential for Large Language Models (LLMs) to address complex problems and enhance robustness and interpretability. Despite the flux of research on developing advanced reasoning approaches, systematically analyzing the diverse LLMs and reasoning strategies in generating reasoning chains remains a significant challenge. The difficulties stem from the lack of two key elements: (1) an automatic method for evaluating the generated reasoning chains on different tasks, and (2) a unified formalism and implementation of the diverse reasoning approaches for systematic comparison. This paper aims to close the gap: (1) We introduce AutoRace for fully automated reasoning chain evaluation. Existing metrics rely on expensive human annotations or pre-defined LLM prompts not adaptable to different tasks. In contrast, AutoRace automatically creates detailed evaluation criteria tailored for each task, and uses GPT-4 for accurate evaluation following the criteria. (2) We develop LLM Reasoners, a library for standardized modular implementation of existing and new reasoning algorithms, under a unified formulation of the search, reward, and world model components. With the new evaluation and library, (3) we conduct extensive study of different reasoning approaches (e.g., CoT, ToT, RAP). The analysis reveals interesting findings about different factors contributing to reasoning, including the reward-guidance, breadth-vs-depth in search, world model, and prompt formats, etc.
PromptAgent: Strategic Planning with Language Models Enables Expert-level Prompt Optimization

Xinyuan Wang, Chenxi Li, Zhen Wang, and 6 more authors

[ICLR 2024] The Twelfth International Conference on Learning Representations, 2024

Abs PDF Code Poster

Highly effective, task-specific prompts are often heavily engineered by experts to integrate detailed instructions and domain insights based on a deep understanding of both instincts of large language models (LLMs) and the intricacies of the target task. However, automating the generation of such expert-level prompts remains elusive. Existing prompt optimization methods tend to overlook the depth of domain knowledge and struggle to efficiently explore the vast space of expert-level prompts. Addressing this, we present PromptAgent, an optimization method that autonomously crafts prompts equivalent in quality to those handcrafted by experts. At its core, PromptAgent views prompt optimization as a strategic planning problem and employs a principled planning algorithm, rooted in Monte Carlo tree search, to strategically navigate the expert-level prompt space. Inspired by human-like trial-and-error exploration, PromptAgent induces precise expert-level insights and in-depth instructions by reflecting on model errors and generating constructive error feedback. Such a novel framework allows the agent to iteratively examine intermediate prompts (states), refine them based on error feedbacks (actions), simulate future rewards, and search for high-reward paths leading to expert prompts. We apply PromptAgent to 12 tasks spanning three practical domains: BIG-Bench Hard (BBH), as well as domain-specific and general NLP tasks, showing it significantly outperforms strong Chain-of-Thought and recent prompt optimization baselines. Extensive analyses emphasize its capability to craft expert-level, detailed, and domain-insightful prompts with great efficiency and generalizability.