The "HKUST Industry Engagement Day Plus" was held on July 10, 2024. As a co-organizer, ACCESS showcased the latest technologies through live demonstrations at the event. In addition, ACCESS’s Principal Researcher, Mr. Mark MOK, delivered a lecture on the topic of "Technology Transfer".
The event was centered around the theme "Transformation to Edge AI 2.0," aiming to foster collaboration among innovators, investors, and industry stakeholders, as well as highlight the latest advancements in this field.
We are excited to highlight in the event two of our research achievements through live demonstrations:
(1) High Energy-efficient Edge Chips & Systems for Large AI Models
The research project utilizes innovative software-hardware co-design to compress large-scale AI models efficiently at the edge. By integrating hardware design architectures and heterogeneous circuits, our solution incorporates Computing In-Memory modules for efficient edge inference computing. We leverage chiplet and high-speed board-level interconnection technologies to offer flexible, energy-efficient, and cost-effective computing systems for diverse edge applications using large AI models.
(2) Hardware Accelerator for Financial Computing
Our financial computing hardware accelerator (AC-CVXPY) utilizes field programmable gate array (FPGA) and application specific integrated circuit (ASIC) to provide the adaptability of a general software solver and enhanced speed through finance-specific hardware architecture.
In the lecture, Mr. MOK delved deeper into the challenges of bringing AI to the edge and the transformative potential of the center's novel platform. He emphasized the need to move beyond the limitations of existing GPU-based edge devices and provide a scalable and adaptable solution that unlocks the full potential of AI at the edge - a critical step towards realizing the vision of Edge AI 2.0.
The showcased leading-edge technologies and research achievements affirm ACCESS's driving role in the forefront of AI development, inspiring the future of intelligent computing at the edge through cross-domain collaboration.
