Latent Action Control for Reasoning-Guided Unified Image Generation
Makes reasoning actionable inside a unified image generator through role-structured latent actions for planning, drafting, diagnosis, and refinement.
Generative AI · Image Generation · Visual Reasoning
About
I am an MPhil student in the Data Science and Analytics Thrust, Information Hub, at The Hong Kong University of Science and Technology (Guangzhou), advised by Prof. Lei Zhu.
My research focuses on image generation and generative multimodal models. I am particularly interested in making visual understanding actionable for generation through unified understanding-generation models, latent reasoning and control, and reward/evaluation models for generative systems.
News
2026.05Latent Action Control, my first-author work on reasoning-guided unified image generation, is under review.
2026.05GenEvolve, our self-evolving image-generation agent framework, is under review.
2026.02PosterReward was accepted to CVPR 2026 (CCF A).
2025.09Started my MPhil study at HKUST (Guangzhou).
Research Interests
Image Generation
High-quality and controllable generative models for visual content creation.
Unified Understanding & Generation
Shared multimodal models that connect visual understanding with image synthesis.
Latent Visual Reasoning
Planning, diagnosis, and refinement through latent actions inside generative models.
Reward Modeling & Evaluation
Reliable preference modeling and evaluation for image-generation systems.
Selected Publications
View allGenEvolve: Self-Evolving Image Generation Agents via Tool-Orchestrated Visual Experience Distillation
A self-evolving image-generation agent that distills structured visual experience from tool-orchestrated generation trajectories.
ROAD: Rule-Grounded Context-Aware Open-World Driver Anomaly Detection
PosterReward: Unlocking Accurate Evaluation for High-Quality Graphic Design Generation
A reward model and evaluation benchmark designed for typography-, layout-, and aesthetics-aware assessment of graphic design generation.
Education
The Hong Kong University of Science and Technology (Guangzhou)
MPhil in Data Science and Analytics · Information Hub
Advisor: Prof. Lei Zhu
Xinjiang University
B.Eng. in Computer Science and Technology · School of Intelligence Science and Technology (School of Future Technology)
Excellent Graduate · Top 12% in major
Selected Honors
View all- Kaggle Bronze Medal — RSNA 2024 Lumbar Spine Degenerative Classification, Top 6%.
- Academic Competitions — First Prize, 14th & 15th National University Students Mathematics Competition in Xinjiang; Third Prize, “Huajiao Cup” National Finals.
- Autonomous Regional People's Government Encouragement Scholarship.
- Outstanding Graduate — Xinjiang University.