Breaking Barriers in Sensory AI: Yang Chai's Energy-Efficient Innovations
Breaking the Wall of Efficient Sensory AI Systems
Winner Interview 2024: Engineering & Technology
Yang Chai is revolutionizing sensory AI by overcoming critical limitations in power consumption, latency, and memory. His pioneering research develops novel hardware architectures and optimization techniques, enabling the deployment of advanced sensory AI systems in mobile devices, IoT sensors, and edge computing, transforming applications in smart cities, autonomous vehicles, and industrial automation.
Which wall does your research or project break?
My research is focused on breaking through the significant energy efficiency limitations that currently constrain sensory AI systems. The key walls we are trying to overcome are the Power Consumption Wall, the Latency Wall, and the Memory Wall.
The Power Consumption Wall:
Current sensory AI systems, such as those used for computer vision and audio processing, require immense computational resources and power, making them energy-intensive and challenging to deploy in resource-constrained environments. Our research aims to develop novel hardware architectures that can drastically reduce the power requirements of these sensory AI pipelines, enabling their widespread adoption in a variety of applications, from mobile devices to IoT sensors and edge computing platforms.
The Latency Wall:
Conventional sensory AI systems often introduce unacceptable levels of latency, which can be a critical limitation for real-time, low-latency applications. Whether it's autonomous vehicles that need to make split-second decisions or industrial automation systems that require immediate responses, the ability to process sensory inputs with minimal latency is paramount. Our work focuses on optimizing the computational efficiency and parallelism of sensory AI by exploring specialized hardware accelerators that can deliver low-latency inference capabilities.
The Memory Wall:
Sensory data processing, particularly for tasks like high-resolution vision and audio analysis, requires substantial memory footprints. This can be a significant obstacle for the deployment of these AI systems on embedded and edge devices, which often have limited on-chip memory. Our research explores innovative memory management techniques, data compression algorithms, and hardware-software co-design approaches to overcome the memory limitations and enable the execution of complex sensory AI models on resource-constrained platforms.
By addressing these key walls of energy efficiency, latency, and memory constraints, our research aims to unlock the full potential of sensory AI systems, paving the way for their widespread adoption in a diverse range of applications, from smart cities and autonomous vehicles to industrial automation and Internet of Things (IoT) devices.
What are the three main goals of your research or project?
The primary goals of our research project are to: 1) Drastically reduce the power consumption of sensory AI systems, 2) Minimize the end-to-end latency of sensory AI processing, and 3) Enable efficient deployment of sensory AI models on resource-constrained platforms by overcoming memory limitations.
Goal 1: Reduce Power Consumption of Sensory AI
Current sensory AI systems, such as those used for computer vision and audio processing, are highly energy-intensive due to their computational complexity. Our work explores novel hardware architectures to address this "Power Consumption Wall." We are developing specialized processing units, novel data representation techniques, and power-efficient circuit designs to dramatically lower the power requirements of these sensory AI models without sacrificing their accuracy or performance.
Goal 2: Minimize Latency of Sensory AI Processing
Many time-critical applications, such as autonomous vehicles, robotics, and industrial automation, require sensory AI systems to process inputs and generate outputs with minimal latency. Our research focuses on optimizing the computational efficiency and parallelism of sensory AI algorithms, as well as exploring specialized hardware accelerators that can deliver low-latency inference capabilities. By addressing the "Latency Wall," we aim to enable the use of sensory AI in a wide range of real-time, mission-critical systems.
Goal 3: Enable Efficient Deployment of Sensory AI
Sensory data processing, particularly for high-resolution vision and audio tasks, often requires large memory footprints that can be a significant obstacle for deploying these AI models on embedded and edge devices with limited on-chip memory. To overcome this "Memory Wall," our innovations in model compression, quantization, and specialized memory architectures aim to enable the efficient deployment of complex sensory AI models on resource-constrained platforms. This will unlock the potential of sensory AI in a diverse range of applications, from smart cities and IoT devices to industrial automation.
By addressing these three key goals, our research project seeks to unlock the full potential of sensory AI systems, paving the way for their widespread adoption across a variety of industries and domains.
What advice would you give to young scientists or students interested in pursuing a career in research, or to your younger self starting in science?
A few years ago, I had the privilege of attending a science event where I had the opportunity to ask a Nobel Laureate how they achieved high-impact research. The Laureate's answer was concise, yet profoundly insightful: 'Curiosity'. That single word encapsulates the essence of what drives groundbreaking scientific discoveries and research. Curiosity - the innate human desire to understand the world around us, to explore the unknown, and to uncover the mysteries of the universe - is the bedrock upon which transformative research is built. My advice to the young generation is to cultivate a deep and genuine curiosity about the world around you.
There are a few other attributes that are also critical for pursing career in scientific research. Identifying and solving complex problems usually take a very long time. Failures and inherent uncertainty are a natural part of the research process. Persistence, resilience, and a willingness to experiment are key attributes for successful researchers during this long process. In addition, many of the most impactful scientific breakthroughs often emerge from interdisciplinary cooperation and diverse perspectives. Cultivating a collaborative mindset and fostering effective communication skills can convey your ideas, findings, and insights to both scientific and non-scientific audiences.
What inspired you to be in the profession you are today?
My advisors and mentors inspired me work towards engineering and technology research.
What impact does your research or project have on society?
The in-sensor computing strategy inspired the development of enhancing decision-making and situational awareness, strengthening privacy and security, revolutionizing intelligent automation, etc. Semiconductor Industry Association has included this paradigm in the Decadal Plan for Semiconductors. His research work on in-sensor visual adaptation has been highlighted in IEEE Spectrum entitled “Realistic retinas make better bionic eyes” in March 2022. This work has been also highlighted in Nature, 2022, 602, 364, emphasizing that “Smart optical sensors inspired by the human retina could improve artificial-vision systems used in autonomous vehicles and industrial manufacturing”. His work was recognized as one of the Top 10 Hong Kong Innovation Technology News in 2022.
What is one surprising fact about your research or project that people might not know?
One surprising fact about our project that people might not know is our exploration of biomimicry and biological inspiration as a key approach. While the project is primarily focused on developing advanced computational and algorithmic techniques for sensory AI systems, we have been closely studying the remarkable capabilities of biological sensory systems found in nature. The goal is to learn from evolution's highly optimized solutions and incorporate some of those principles and mechanisms into the design of their artificial sensory processing pipelines. Studying and emulating nature's proven solutions can provide valuable insights and design principles that may lead to breakthroughs in this field.
What’s the most exciting moment you've experienced over the course of your research or project?
The most exciting moment I've experienced over the course of our research project was the successful demonstration of our sensory processing architecture in an adaptive manner. After months of rigorous research, design, and prototyping, we were finally able to put our ideas into practice and see the system in action.
