Welcome to my GitHub profile! I am an Undergraduate Computer Science student at Sun Yat-sen University (SYSU) and a Research Intern at Microsoft Research Asia (MSRA).

My long-term goal is to build a unified world model that seamlessly integrates understanding and generation across modalities and agents.

📢 I am seeking PhD opportunities for Fall 2027. If you are interested, please contact me directly!

• Education: B.Eng. in Computer Science, Sun Yat-sen University (Sept 2023 - Present).
• Research Labs:
  • Microsoft Research Asia (MSRA), Shanghai ML Group.
  • HCP Lab, Sun Yat-sen University.
• Role: Reviewer for CVPR 2026, AAAI 2026, ICLR 2026.
• Interests beyond tech: Philosophy (Kant, Nietzsche), Literature (Tolstoy, Kafka), and the intersection of humanity and technology. 📚🤔

My research centers on advancing multi-agent collaborative intelligence by integrating Visual Language Models (VLMs) and Large Language Models (LLMs) with Reinforcement Learning (RL).

• Multi-Agent Systems: Knowledge-aware coordination, Bayesian bandits, Game-theoretic uncertainty trading.
• Policy Optimization: PPO, MAB, GRPO, and MAPPO-like RL frameworks.
• Generative Models: Video generation (Hierarchical VAE), 3D generation, and Unified World Models.
• VLM Architectures: Discretized representations (VQ), Post-training with RFT (LLaVA).

Check out my Google Scholar for the full list.

• [AAAI 2026] Cost-Effective Communication: An Auction-based Method for Language Agent Interaction
  Yijia Fan, Kaitong Cai, et al.
• [AAAI 2026] 3DAlign-DAER: Dynamic Attention Policy and Efficient Retrieval Strategy for Fine-grained 3D-Text Alignment
  Yijia Fan, Jusheng Zhang, Kaitong Cai, et al.
• [NeurIPS 2025] GAM-Agent: Game-Theoretic and Uncertainty-Aware Collaboration for Complex Visual Reasoning
  Jusheng Zhang, Yijia Fan, et al.

• Microsoft Research Asia (Shanghai ML Group) | July 2025 – Present
  • Improving video generation using hierarchical VAE and exploring next-gen discretized VLM projects.
  • Training Unified video generation/understanding models on multi-node clusters (DeepSpeed).
• HCP Lab (SYSU) | July 2024 – Present
  • Exploring long-context LLM processing via diffusion models and curiosity-based game systems for few-label classification.

• Languages: Python (PyTorch), C/C++. 🐍💻
• Frameworks: PyTorch, CUDA, DeepSpeed, PyTorch Lightning. ⚡
• Tools: Linux, LaTeX, Git. 🐧📄

Whether you’re into AI research, philosophy, or literature, feel free to reach out!

📫 Email: fanyj28@mail2.sysu.edu.cn
📍 Location: Guangzhou / Shanghai, China