Anbu Huang

Anbu Huang

Research Scientist, Author, AI Enthusiast, THU

Welcome to Anbu Huang’s HomePage

About me:

Over the past few years, I have dedicated my career to advancing the field of Generative AI, particularly through the development and application of Diffusion models and LLM. These technologies have been at the core of my work, I aim to develop scalable, efficient systems that leverage these technologies for innovative content generation, knowledge transfer, and multimodal applications, while ensuring practical deployment and ethical considerations.

  • Diffusion Models: I have developed multimodal generation systems operating in both RGB and Bayer spaces, enabling high-fidelity synthesis of images, videos, and other visual data. These systems support applications in creative tools, sensor data processing, and augmented reality, with an emphasis on controllability and cross-modal integration.

  • Large Language Models: Based on distributed architectures, I have built knowledge distillation systems for collaborative large-small model frameworks. This involves using server-side large models to provide knowledge transfer—such as data augmentation through synthetic data generation, feature alignment, and chain-of-thought migration—to enhance domain-specific small models, improving their performance, efficiency, and adaptability in real-world scenarios.

In my past work (2014 - 2023), I focused on applied AI, including recommendation systems, federated learning, and AI safety, building scalable solutions that prioritized privacy, security, and user impact.

  • Recommendation Systems: At Tencent, I led a technical team to develop the world’s largest online music recommender system, serving over 100 million users globally. We utilized AI techniques like natural language processing, music information retrieval, and deep neural networks to analyze listening histories and predict user preferences, modeling human emotions and interests.
  • Federated Learning and Privacy Computing: I engaged in theoretical and practical aspects of trustworthy machine learning, participating in the design of the FATE framework, developing a federated vision system, and publishing several important research papers on federated learning .
  • AI Safety and Anti-Spoofing: To address risks from technologies like Deepfakes, I contributed to facial verification systems, including liveness detection and recognition, while developing frameworks to detect biases, adversarial attacks, and ensure ethical AI deployment.