I am Haoxi Li (李昊翕), a Ph.D. student in the Department of Computer Science and Engineering at The Hong Kong University of Science and Technology (HKUST).
My research focuses on reasoning and decision-making in foundation models, particularly large reasoning models, reinforcement learning, and vision-language agents. I am interested in developing intelligent systems that can reason efficiently, evaluate their own decisions, and continually improve through interaction with their environments.
You can find my publications on Google Scholar.
🔬 Research Interests
- Large Reasoning Models: Efficient reasoning, metacognition, and reliable model behavior.
- Agentic Reinforcement Learning: Self-exploration, test-time adaptation, and adaptive decision-making.
- Vision-Language Agents: World modeling, multimodal reasoning, and curiosity-driven exploration.
🔥 News
- 2026.07: Our paper on visual-linguistic curiosity (GLANCE) was published at ICML 2026.
- 2026.07: Our papers Think How to Think and TTVS were published at ACL 2026 and Findings of ACL 2026, respectively.
- 2025.04: Our paper CausM was accepted at ICLR 2025.
📝 Selected Publications
2026
What You Think is What You See: Driving Exploration in VLM Agents via Visual-Linguistic Curiosity
Haoxi Li, Qinglin Hou, Jianfei Ma, Jinxiang Lai, Tao Han, Sikai Bai, Jingcai Guo, Jie Zhang, Song Guo
International Conference on Machine Learning (ICML), 2026
A curiosity-driven exploration framework that connects linguistic world modeling with visual feedback in VLM agents.
Yongjiang Liu, Haoxi Li, Xiaosong Ma, Jie Zhang, Song Guo
Annual Meeting of the Association for Computational Linguistics (ACL), 2026
Improving reasoning efficiency by enabling language models to recognize task difficulty and reduce redundant reasoning.
TTVS: Boosting Self-Exploring Reinforcement Learning via Test-time Variational Synthesis
Sikai Bai, Haoxi Li, Jie Zhang, Yongjiang Liu, Song Guo
Findings of the Association for Computational Linguistics (ACL), 2026
A test-time self-exploration framework that improves reasoning models through variational synthesis and unlabeled queries.
2025
CausM: Causally Motivated Sycophancy Mitigation for Large Language Models
Haoxi Li, Xueyang Tang, Jie Zhang, Song Guo, Sikai Bai, Peiran Dong, Yue Yu
International Conference on Learning Representations (ICLR), 2025
A causally motivated framework for mitigating sycophantic behavior and improving the reliability of large language models.
CoRE: Enhancing Metacognition with Label-free Self-evaluation in LRMs
Haoxi Li, Sikai Bai, Jie Zhang, Song Guo
arXiv Preprint, 2025
A latent-representation-based approach to self-evaluating reasoning trajectories without external labels.
📖 Education
- Ph.D. Student, Department of Computer Science and Engineering, The Hong Kong University of Science and Technology (HKUST).