Qiang Lyu

I am a direct-entry PhD student in Computer Science at the University of Chinese Academy of Sciences (UCAS), where I work on generative models, multimodal understanding, and reinforcement learning.

My recent work focuses on reliable video reasoning, text-to-image generation, cross-domain image composition, and efficient video generation. I enjoy turning research questions into reproducible experiments and practical systems.

Some of my technical posts include:

• A Complete Mathematical Derivation of DDPM: Denoising Diffusion Probabilistic Models
• Flow-GRPO: Training Flow Matching Models via Online RL
• Diffusion Models: From SDEs and ODEs to One-Step Generation

I have worked on video generation, multimodal reasoning, and generative-model optimization during research internships at Alibaba Cloud and Youku, and algorithm internships at RightBrainAI and Megvii. I expect to complete my PhD in June 2027.

You can find my work on Google Scholar and GitHub.