Hong Kong Generative AI Research & Development Center

Focusing on research and application in generative AI by training Hong Kong's first locally developed AI foundation model

The AI foundation model is a key infrastructure for advancing the field of artificial intelligence. Its powerful cross-disciplinary applicability, collaborative resource capability, and boundless innovative potential profoundly transform traditional methods of technology development and application. From legal consultation to artistic creation, daily writing to medical diagnosis, the features of foundation models can be used in various scenarios and industries, effectively solving numerous practical problems in society.

HKGAI team members

HKGAI Center Director Prof. Yike Guo and researchers

HKGAI V1 LLM Press Conference

Given the vast opportunities presented by artificial intelligence, the inception of HKGAI marks a significant milestone. HKGAI focuses on the research and development of generative artificial intelligence technologies. We aim to benefit Hong Kong by building and open-source our first locally developed foundation model. Based on the development of the foundation model, the speed of progress in technology and the development of applications will see significant acceleration. It also actively contributes to Hong Kong's role in driving economic prosperity within the Greater Bay Area.

HKGAI has established eight work packages, including research and development of foundation models, as well as the development and construction of foundation models across various industries.

• Foundation Model Development: Foundation Model Development, emphasizing practicality, as well as the maintainability and scalability of the system, making it an open-source development platform for large communities, achieving flywheel development.
• Foundation Model Research: The Foundation Model Research will focus on developing foundation models for various target fields, investigating the impact of different training methods, training data, and model architectures on the foundational model.
• Infrastructure for Foundation Model: The Infrastructure for Foundation Model focuses on developing an easy-to-use and high-performance AI infrastructure for the whole life cycle of implementing the HKGAI on developing, maintaining, and applying the Foundation Models.
• Data Engineering for Foundation Model: The Data Engineering for Foundation Models in this research project will focus on delivering practical solutions for acquiring and managing a large data set for training the foundation model, as well as conducting quality assessments of acquired data, and constructing high-quality data annotations for model tuning and domain-specific model induction.
• Medical Foundation Model System: The Medical Foundation Model System will focus on conducting research on how to effectively train and define a large-scale Medical Foundation Model with billions of trainable parameters through adapting the HKGAI based on multimodal clinical data. More importantly, the Medical Foundation Model System is ethical, complies with related standards and effectively protects the privacy of medical data.
• Legal Foundation Model System: The Legal Foundation Model System will integrate legal knowledge with foundation models for enhanced understanding and efficient content generation. We emphasize practical use, self-reinforcement, and feedback from legal professionals.
• Multilingual Foundation Model System: The Multilingual Foundation Model System will solve the problem of how to transform multimodal information such as text, audio, and video into the unified semantic embedding space for optimal representation. We aim to ensure that artificial intelligence-generated results align with human cognition and aesthetic judgments. The outcomes are presented as artistic expressions in everyday life, enriching human’s aesthetic evaluation.
• Multimodal Foundation Model: The Multimodal Foundation Model will focus on developing a foundation model for speech based on discrete speech units. We seek to develop data augmentation techniques to address data sparsity, particularly in Cantonese speech.

In addition, HKGAI focuses on cultivating AI talents and ecosystem in Hong Kong, conducting research on ethics, safety, and governance in AI technology and applications. These efforts aim to enhance Hong Kong's global influence in AI research and application.