Generative Model
Generative capabilities, reinforced reasoning, and world modeling.
I’m Yuwei. My research spans Generative Model, Agent System, and Simulation — connecting capabilities, interactions, and environments.
Explore my research ↘Research, open-source systems, and the curiosity that connects them.
I study the connections between models that generate, agents that reason and act, and environments in which behavior can be explored.
Generative capabilities, reinforced reasoning, and world modeling.
Reasoning, action, and interaction — connected into workflows and larger systems.
Social behavior and urban activity, studied through agents in an environment.
A complete list of thirteen distinct works, with publication venues and author roles kept visible.
Generative agents · Human behavior
AgentSociety preprint · arXiv v2 · 2026
OpenCity · Urban simulation
OpenCity preprint · arXiv v1 · 2024; not final KDD version
Embodied AI · World models
Recommender systems · Agent benchmarks
arXiv v2 (2025)
Reinforced reasoning · Language models
arXiv v3 (2025)
Social agents · Workflow design
arXiv v1 (2025) preprint
Social experiments · LLM agents
arXiv PDF header states Published as a conference paper at COLM 2025.
Multi-agent systems · Simulation
Official ACL 2025 Industry proceedings PDF
User modeling · Recommendation
arXiv2502.18754v1,2025-02-26; author preprint of same publication family
Social simulation · International perceptions
arXiv2508.08837v3,2026-03-10
Distributed inference · Dynamic scheduling
Compiler optimization · Heterogeneous systems
Author-hosted IEEE published paper, vol73 no7 July2024; current version2024-06-11
Human mobility · Agent benchmarks
Figures open at a larger size with source links. Preprint figures are identified separately from final publication metadata. Preprint and final versions are grouped as one work.
Research questions become systems to inspect, use, and build upon. AgentSociety leads this collection.
Generative agents, realistic environments, and social experiments in one research platform.
Co-first author · iFuture 2026
Explore the research ↗Connecting source excerpts, assertions, and conflicting evidence into a research knowledge system.
Multi-agent planning, writing, review, and LaTeX production for academic work.
A scalable LLM-agent framework for urban activity and large-scale urban segregation simulation.
The photographs live together in one changing composition at the top of the page: a wider perspective, a quieter moment, and another kind of expression.
Revisit the photographs ↑