People of ACM - Tim Cheng

May 7, 2026

Kwang-Ting (Tim) Cheng is the Vice-President for Research and Development and a Chair Professor at The Hong Kong University of Science and Technology (HKUST). His research interests focus on electronic design automation and software-hardware co-design for computing chips and systems. He has extensive experience in fostering cross-disciplinary research collaborations and has also made significant contributions to computer vision and medical image analysis. His work has earned him several accolades, including the ACM SIGDA Pioneering Achievement Award, the Pan Wen Yuan Foundation Award for Outstanding Research, CCF Overseas Outstanding Contribution Award, and over a dozen best paper awards at major conferences and journals. Additionally, he has successfully translated several of his inventions into commercial products.

Cheng was recently recognized as an ACM Fellow for his contributions to design automation and software-hardware co-design of electronic circuits and computing systems. He is also a Fellow of the Hong Kong Academy of Engineering Sciences and a Fellow of the School of Engineering at the University of Tokyo.

Which of your research areas are you devoting the most of your time to right now?

My research team and I are working on two key areas. First, we’re developing highly energy-efficient AI accelerator architectures and co-design methodologies for AI edge inference at scale. At the same time, we’re developing predictive vision models for medical image analysis and multi-modal models for health applications. The goal of combining these two initiatives is to gain better insights and address the demanding computational requirements of edge and physical AI through practical, high-impact applications.

What are some of the most exciting trends in design automation for computer systems?

Computing systems for AI applications face the pressing constraint of the “energy wall” (the point at which AI is limited by the electricity supply). This constraint elevates energy efficiency to a paramount, first-class metric, acting as the greatest catalyst for innovations in the design of AI computing systems. This catalyst is driving exciting trends, particularly in co-design and co-optimization across the entire stack. The cross-layer co-design/-optimization approaches span from the application layer and algorithm/model layer through to hardware architectures and even fabrication and packaging technologies, with the ultimate goal of achieving orders-of-magnitude gains in performance-per-watt.

In one of your most cited papers, “Fast Human Detection Using a Cascade of Histograms of Oriented Gradients,” you (along with co-authors Qiang Zhu, Mei-Chen Yeh, and Shai Avidan) presented a new approach to achieve fast and accurate human detection (an outgrowth of face detection algorithms). Will you discuss how this was an improvement of the state-of-the-art at the time? Your paper was written in 2006. How are computers doing with human detection in 2026?

Before our 2006 paper, human detection was a significantly challenging task, and researchers typically faced a substantial trade-off between accuracy and speed. Many existing methods were either accurate but slow, or fast but prone to false positives and missed detections. Our approach, the cascade of Histograms of Oriented Gradients (HOG), represented a substantial improvement by achieving both high accuracy and real-time speeds. This was a major leap forward in 2006, enabling a range of applications that were previously not feasible due to computational limitations.

This breakthrough was achieved by effectively and synergistically combining two key components: a powerful and robust feature descriptor, HOG, which captured the shape and appearance of humans well with an efficient detection framework based on the cascade of classifiers. The cascade allowed us to quickly discard obvious non-human regions, focusing computational resources only on more promising areas, thus achieving the desired speed without sacrificing accuracy.

The landscape of human detection in 2026 is dramatically different, largely due to the dominance of deep learning. Instead of relying on hand-crafted features, deep learning models learn features directly from raw image data. New approaches are capable of discovering highly complex and abstract representations that are far more powerful and discriminative than what was possible with earlier methods. The current state-of-the-art leverages the power of end-to-end deep learning to achieve unprecedented levels of accuracy, robustness, and even speed on much more complex tasks, such as detecting humans in diverse and occluded scenarios, estimating their pose, and even understanding their actions.

As the Director of HKUST’s AI Chip Center for Emerging Smart Systems (ACCESS), what are the most important ways AI chip design must improve? How should we be developing the next generation of talent for the global AI chip market?

I see critical improvements needed in AI chip design technologies, particularly for edge and physical AI, alongside a strategic approach to talent development. The important aspects driving AI chip and system design include:

• Ubiquitous, Efficient Edge AI: We need chips and systems that are not just powerful but ultra-low power, cost-effective, and miniaturized for widespread embedding.
• Deep Cross-Layer Co-Design: cross-layer co-design and co-optimization across applications, AI models, and chip/memory architectures is paramount. This holistic approach requires rethinking traditional designs, exploring innovations in architecture and electronic design automation (EDA) solutions to break through current bottlenecks.
• Scalability and Adaptability: AI chip designs must be inherently scalable and flexible to accommodate fast-evolving AI paradigms and models. This means prioritizing programmability, reconfigurability, and reusability to adapt to future algorithms without complete hardware redesigns.
• Trustworthiness and Security: AI chips must incorporate robust hardware-level security features and ensure fault tolerance and reliability for safe and secure operation.

To address the needs of the fast-evolving global AI chip market, we must cultivate a multidisciplinary talent pool. There is a strong and growing demand for hybrid expertise and developing individuals with skills and knowledge of both "hardware-aware" AI and "AI-aware" hardware. This means training students with the expertise to work in both AI algorithms and chip design/computer architecture alongside mastering modern EDA tools.

You have written about making Hong Kong an innovation hub on par with Silicon Valley. What has Hong Kong done to start on this path? What new initiatives are planned toward this goal?

Hong Kong has been actively laying the groundwork to position itself as a leading innovation hub, with the consensus that innovation and technology being one of its core economic pillars. Its inherent internationalization and role as a bridge between East and West provides a natural advantage for fostering global collaboration. Furthermore, this region’s commitment to higher education and talent nurturing creates a fertile ground for research, development, and innovation.

Hong Kong has launched several major initiatives:

• InnoHK (Research Clusters) Program: This flagship program aims to establish world-class, interdisciplinary research laboratories and centers by attracting leading international universities, research institutions, and companies to collaborate with local universities and talent. The focus is on key areas like healthcare technology, AI and robotics, fintech, and new energy and environmental technology, demonstrating a clear strategic direction for innovation.
• Startup Ecosystem Development: Beyond large-scale research centers, there is a concerted effort to cultivate a vibrant startup ecosystem. This includes supporting incubators and accelerators with substantial and systematic financial backing.
• Talent Attraction and Retention: Hong Kong is implementing policies to attract and retain top global talent, including streamlined visa processes for skilled professionals, researchers, and entrepreneurs, as well as initiatives to encourage local talent development through specialized programs.

These initiatives collectively aim to create an open, collaborative environment where cutting-edge research can flourish, innovative businesses can emerge, and Hong Kong can secure its place as a global innovation powerhouse.