REINFORCEMENT LEARNING & MULTIMODAL AI
Exploring how machines
perceive and create.
I am a master’s student at Peking University. My research focuses on reinforcement learning, agentic RL, and multimodal learning. I am interested in how models learn to reason, use tools, and understand and generate multimodal content.
M.S. Student in Computer Science · Peking University
B.S. in Computer Science · Sichuan University
BUILDING IN THE OPEN
Contributions through internships
at Tencent Hunyuan and Ant Group.
As a main contributor to UniRL, a reinforcement learning framework for unified multimodal models, I primarily developed the prompt enhancement (PE) trainer and agentic RL components, supporting image-feedback-driven prompt optimization and multi-turn tool-use training.
Explore repository ↗
I contributed to DLRover, an automatic distributed deep learning system for efficient and reliable training. My work focused on designing and implementing Flash-checkpoint, along with Auto Accelerate experiments and analysis for large-scale recommendation model training.
Explore repository ↗IDEAS INTO PRACTICE
Generative models, perception,
and the systems behind them.
Conference on Neural Information Processing Systems (NeurIPS), 2025. (CCF-A)
European Conference on Computer Vision (ECCV), 2026.
Proceedings of the VLDB Endowment (PVLDB), 2024. (CCF-A)
* Equal contribution.
THE PATH SO FAR
M.S. in Computer Science and Technology
B.S. in Computer Science and Technology
RESEARCH EXPERIENCE
RESEARCH INTERN · FROM MAR 2025
For AlignedGen, I co-led the work as an equal-contribution first author, designing the core method, implementing the pipeline, and conducting the main experiments.
In collaboration with THU IVG Lab, I led the core method design, end-to-end implementation, and major experiments for DDAVS on audio-visual segmentation.
A FOUNDATION IN PROBLEM SOLVING
LET’S CONNECT
