Zhihao Yang

yangzhihao25 [at] mails.ucas.ac.cn

Hello! Welcome to my homepage.

I am a second-year Ph.D. student in Computer Science at the University of Chinese Academy of Sciences (UCAS), advised by Prof. Xianpei Han. Prior to this, I obtained my B.Eng. degree in Computer Science from Tongji University. I am interested in Game & Agent. Previously, I was fortunate to be advised by Prof. Min Yang and Prof. Yujun Cai.

I enjoy basketball and rock climbing. My favorite team is the Golden State Warriors. I am a huge fan of Stephen Curry!

Publications

First Author * equal contribution; † corresponding author
	Scaling in Games: Continued Pre-Training for Embodied Agents in Diverse Virtual Worlds Kuan Zhang^, Yukun Chen^, Zhihao Yang^*, Yue Su, Xiangnan Wu, Zirong Chen, Run Luo, Jinkun Hou, Tao Tan, Yinhe Zheng^†, Yiming Li^† arXiv, 2026 [paper] [project page] [github] [models] [demo] How data, world diversity, and model size scale continued pre-training for embodied agents.
	Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal Capability Kuan Zhang^, Yukun Chen^, Zhihao Yang^*, Yue Su, Xiangnan Wu, Zirong Chen, Run Luo, Tao Tan, Yinhe Zheng^†, Yiming Li^† arXiv, 2026 [paper] [project page] [github] [dataset] [demo] A video benchmark for perceiving events, reading state updates, and tracking an evolving world.
	PIVOT: Progressive Intervention via Online Teacher Correction Zhihao Yang^, Boyuan Zheng^, Puhui Tan^*, Yukun Chen, Kuan Zhang, Longze Chen, Jiaming Li, Renke Shan, Lu Wang^†, Xianpei Han^† arXiv, 2026 A teacher-assisted RLVR method that inserts teacher corrections into student rollouts and applies token-level KL supervision, nearly doubling mean avg@8 on hard logic games.
	Evolving to Match Capabilities: Improving Harness Self-Evolution via Skill Adaptation Zhihao Yang^, Haojin Wang^, Juncheng Wu^*, Kuan Zhang, Yukun Chen, Run Luo, Yaojie Lu, Yuyin Zhou, Lu Wang^†, Xianpei Han^† arXiv, 2026 [paper] [project page] [github] [X] A meta-skill that adapts harness guidance to the executor's capability, so self-evolution helps both weak and strong agents without updating model parameters.
	Evade: Multimodal benchmark for evasive content detection in e-commerce applications Ancheng Xu^, Zhihao Yang^, Jingpeng Li, Guanghu Yuan, Longze Chen, Liang Yan, Jiehui Zhou, Zhen Qin, Hengyun Chang, Hamid Alinejad-Rokny, Bo Zheng, Min Yang. SIGIR, 2026 (CCF-A) [project page] [arXiv] EVADE-Bench introduces a multimodal benchmark to evaluate evasive content detection and provides a strong foundation for developing more robust detection systems.
	Structuring Reasoning for Complex Rules Beyond Flat Representations Zhihao Yang^, Ancheng Xu^, Jingpeng Li, Liang Yan, Jiehui Zhou, Zhen Qin, Hengyu Chang, Yukun Chen, Longze Chen, Ahmadreza Argha, Hamid Alinejad-Rokny, Minghuan Tan, Yujun Cai, Min Yang arXiv, 2025 [arXiv] An approach which equips LLMs with a structured template to methodically gather and verify evidence within complex rule systems, drastically improving reasoning fidelity.
Co-authored Papers
	LLM2Cert: LLM Synthesis of Certificates with SMT/Lean Verification Zhipeng Cao^, Peiyang Song^, Nikolaus Vertovec, Weiru Cui, Zhihao Yang, Mingkang Liu, Luxia Lin, Bai Xue^†, Alessandro Abate^† arXiv, 2026 [paper] [project page] [github] [dataset] [X] An agent framework that synthesizes Lyapunov and barrier certificates for dynamical systems and verifies them with SMT and Lean, without accepting solver relaxations.
	FlowPIE: Test-Time Scientific Idea Evolution with Flow-Guided Literature Exploration Qiyao Wang, Hongbo Wang, Longze Chen, Zhihao Yang, Guhong Chen, Hamid Alinejad-Rokny, Hui Li, Yuan Lin, Min Yang. arXiv, 2026 [paper] [project page] [github] [data] FlowPIE tightly couples literature retrieval with idea generation through flow-guided search and test-time evolution, producing research ideas that are more novel, diverse, and feasible than strong LLM- and agent-based baselines.
	RuCL: Stratified Rubric-Based Curriculum Learning for Multimodal Large Language Model Reasoning Yukun Chen, Jiaming Li, Longze Chen, Ze Gong, Jingpeng Li, Zhen Qin, Hengyu Chang, Ancheng Xu, Zhihao Yang, Hamid Alinejad-Rokny, Qiang Qu, Bo Zheng, Min Yang. ICML, 2026 (CCF-A) [arXiv] RuCL reframes curriculum learning around reward design by stratifying rubrics according to model competence, then dynamically reweighting them to progressively improve multimodal reasoning from perception to higher-order logic.