I am a Ph.D. student at Wuhan University (WHU), advised by Prof. Lefei Zhang, with an expected graduation date of December 2027. I am currently a research intern with the LongCat Team (Agent) at Meituan through the Beidou Program, working on reinforcement learning for large language models and agents. I have also been fortunate to work with Dr. Sen Zhang, Dr. Liang Ding, and Prof. Dacheng Tao. Previously, I received my bachelor’s degree in Mathematics and Applied Mathematics from the University of Electronic Science and Technology of China (UESTC) in 2022, advised by Prof. Xile Zhao. I am sincerely grateful to all my mentors for their invaluable guidance, support, and inspiration throughout my academic journey.
I have authored or co-authored 20 papers, including work published in top-tier international journals and conferences such as IEEE TPAMI, ICML, NeurIPS, CVPR, ICCV, and EMNLP, as well as manuscripts currently under review. My research has received 1,000+ citations on Google Scholar. In addition, I am supported by the Fundamental Research Project for Young Professional from NSFC (国自然博士生基金, PR). I have also received several honors, including the China Remote Sensing Outstanding Achievement First-Class Prize (2025 中国遥感优秀成果一等奖, PR) and the VALSE 2025 Popular Poster Award (VALSE 2025人气海报奖, PR).
My research focuses on understanding and improving the stability of reinforcement learning for large language models through the lens of its underlying mechanisms, progressing from RLHF to RLVR and long-horizon RL. My work also extends to asynchronous RL systems, multi-agent systems, agent harness self-improvement, and safety. Previously, I worked on tensor modeling and computation, high-dimensional image processing, and remote sensing.
📧 I am open to collaboration and welcome inquiries from anyone interested in my research topics. Feel free to reach out via szmyc1@163.com or miaoyuchun@whu.edu.cn.
🔥 News
- 2026.09: 🎉🎉 One paper has been accepted by NeurIPS 2026.
- 2026.05: 🎉🎉 HyperSIGMA was selected as an ESI Hot Paper (Top 0.1%).
- 2025.12: 🎉🎉 I was supported by the Fundamental Research Project for Young Professional from NSFC (国家自然科学基金博士生专项).
- 2025.11: 🎉🎉 HyperSIGMA was selected for the China Remote Sensing Outstanding Achievement First-Class Prize (2025 中国遥感优秀成果一等奖).
- 2025.11: 🎉🎉 HyperSIGMA has been selected as ESI Highly Cited Papers (TOP 1%).
- 2025.10: 🎉🎉 One paper has been accepted by IEEE GRSL.
- 2025.08: 🎉🎉 One paper has been accepted by EMNLP 2025.
- 2025.06: 🎉🎉 HyperSIGMA received the VALSE 2025 Popular Poster Award (VALSE 2025人气海报奖).
- 2025.04: 🎉🎉 One paper has been accepted by ICML 2025.
- 2025.03: 🎉🎉 One paper has been accepted by IEEE TPAMI.
- 2024.09: 🎉🎉 One paper has been accepted by NeurIPS 2024.
- 2024.01: 🎉🎉 One paper has been accepted by IEEE TCI.
- 2023.07: 🎉🎉 One paper has been accepted by ICCV 2023.
- 2023.02: 🎉🎉 One paper has been accepted by CVPR 2023.
- 2021.08: 🎉🎉 One paper has been accepted by IEEE TGRS.
📝 Selected Publications
[1] When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR
First author · Under review · Paper
[2] DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training
Co-first author · NeurIPS 2026 · Deployed in LongCat production training · Paper
[3] The Energy Loss Phenomenon in RLHF: A New Perspective on Mitigating Reward Hacking
First author · ICML 2025 · Paper · Code
[4] InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling
First author · NeurIPS 2024 · 130+ citations · Paper · Code
[5] Information-Theoretic Latent Space Modeling for Robust RLHF under Imperfect Human Feedback
First author · Under review · Paper · Code
[6] HyperSIGMA: Hyperspectral Intelligence Comprehension Foundation Model
Co-first author · IEEE TPAMI 2025 · ESI Hot Paper (Top 0.1%) · 350+ citations · Paper · Code
[7] PluginRSI: Recursive Improvement of Agent Harnesses with Reusable Plugins
Second author · Under review · Paper
[8] Skill1: Unified evolution of skill-augmented agents via reinforcement learning
Fourth author · Under review · Paper
[9] AD-GRPO: Associative Diversity-Aware GRPO for Multimodal Reasoning
Second author · Under review
[10] AMIA: Automatic Masking and Joint Intention Analysis Makes LVLMs Robust Jailbreak Defenders
Second author · EMNLP 2025 · Paper
[11] DDS2M: Self-Supervised Denoising Diffusion Spatio-Spectral Model for Hyperspectral Image Restoration
First author · ICCV 2023 · 110+ citations · Paper · Code
[12] Snapshot Compressive Imaging Using Domain-Factorized Deep Video Prior
First author · IEEE TCI 2023 · Paper · Code
[13] Hyperspectral Denoising Using Unsupervised Disentangled Spatiospectral Deep Priors
First author · IEEE TGRS 2022 · Paper · Code
[14] Uncertainty-Aware Unsupervised Image Deblurring with Deep Residual Prior
Fifth author · CVPR 2023 · Paper · Code
[15] PHDMamba: Progressive Hybrid Mamba for Hyperspectral Image Classification
Fifth author · IEEE GRSL 2025 · Paper · Code
[16] Complex Video Completion Fusing Low-Rank Background and Deep Foreground Priors
Fourth author · IEEE SPL 2022 · Paper
🎖 Honors and Awards
- Fundamental Research Project for Young Professional from NSFC, 2025 December
(主持国家自然科学基金青年学生基础研究项目（博士研究生), Link) - China Remote Sensing Outstanding Achievement First-Class Prize, 2025 December
(中国遥感优秀成果一等奖, Link) - National Scholarship, Ministry of Education of China, 2025 October
(国家奖学金, Top 2% in Wuhan University) - VALSE Popular Poster Award, 2025 June
(VALSE 人气海报奖, 11/398 , Link) - National Scholarship, Ministry of Education of China, 2024 October
(国家奖学金, Top 2% in Wuhan University)
📖 Educations
- Wuhan University, September 2022 – Present (Expected graduation: December 2027)
Ph.D. Student in the School of Computer Science, Wuhan, China.
Supervised by Lefei Zhang. - University of Electronic Science and Technology of China, September 2018 – July 2022
Undergraduate Student in the School of Mathematical Science, Chengdu, China.
Supervised by Xile Zhao.
💻 Internships
- LongCat Team (Agent), Meituan, January 2026 – Present
Research Intern on Agentic RL (Beidou Program), advised by Qi Gu. - A Generative AI Research Startup, April 2023 – October 2025
Research Intern on LLM Alignment, advised by Liang Ding.