TLDR: The Hong Kong Polytechnic University (PolyU) has unveiled significant advancements in Generative AI (GenAI) through its novel ‘Co-GenAI’ collaborative training paradigm. This breakthrough shifts from centralized to decentralized model training, drastically cutting computational costs, enhancing data privacy, and fostering a more inclusive global AI research environment. The initiative aims to position Hong Kong as a leading hub for GenAI innovation.
The Hong Kong Polytechnic University (PolyU) is making waves in the artificial intelligence landscape with groundbreaking advancements in Generative AI (GenAI). The PolyU Academy for Artificial Intelligence (PAAI) has introduced a revolutionary collaborative training paradigm, dubbed ‘Co-GenAI,’ which promises to democratize access to advanced AI research and development.
Traditionally, training large-scale foundation models for GenAI has been an immensely resource-intensive endeavor, often limited to a few organizations with vast computational power. PAAI’s Co-GenAI framework addresses this by transforming the conventional centralized model training into a decentralized approach. This innovative method allows for the merging of separately trained AI models into one robust system, significantly reducing the prohibitive computational costs and breaking down barriers such as GPU monopolies. The result is a more inclusive environment, enabling global institutions and academic researchers to participate in cutting-edge AI research.
Professor Yang Hongxia, Executive Director of PolyU PAAI, emphasized the significance of this shift, stating, ‘Ultra-low-resource foundation model training, combined with efficient model fusion, enables academic researchers worldwide to advance GenAI research through collaborative innovation.’ The team’s research has shown remarkable efficiency gains, reducing video memory usage by approximately 24 percent and cutting training time by 48 percent compared to mainstream BF16 approaches, effectively halving the required computing power for similar results.
Beyond efficiency, Co-GenAI prioritizes data privacy and addresses the challenge of siloed domain knowledge. By allowing models to be trained separately and then merged, it protects sensitive data while still enabling the creation of powerful, comprehensive AI systems. The PAAI team has also provided the first theoretical validation of model fusion, proposing a ‘Model Merging Scaling Law,’ which offers a potential new pathway towards achieving artificial general intelligence (AGI).
PolyU’s commitment extends to practical applications. PAAI has already applied its pipelines to develop domain-specific models, including advanced medical and cancer AI systems, achieving best-in-class results. These specialized AI tools are being developed in collaboration with hospitals, including Queen Elizabeth Hospital in Hong Kong and other mainland institutions, aiming to enhance personalized treatment planning.
Also Read:
- Lingnan University and University of Bologna Co-Host Workshop on AI Governance Technologies, Emphasizing Stronger Oversight for Agentic and Open-Source AI
- MIT’s Project NANDA Forges Path Towards a Decentralized ‘Internet of AI Agents’
Professor Christopher Chao, Senior Vice President (Research and Innovation) at PolyU, highlighted the broader impact of these developments: ‘AI is a key driver for new, quality productive forces. PAAI accelerates AI integration across industries, developing domain-specific models that enhance PolyU’s leadership and position Hong Kong as a global hub for GenAI.’ The project receives substantial support, funded under the Theme-based Research Scheme 2025/26 by the Research Grants Council, the One-plus Scheme under the HKSAR Innovation and Technology Commission, and Cyberport’s AI Subsidy Scheme. This initiative marks a pivotal step for Hong Kong in global AI innovation, fostering both the democratization and industrial adoption of AI technologies.
