Why this research is valuable

Green AI models are meticulously crafted from the ground up to prioritize energy efficiency. By employing techniques like network pruning and knowledge distillation, we ensure that our AI models are not only powerful but also incredibly lean, trimming any unnecessary computational fat. This is complemented by a lightweight machine learning protocol that slashes energy consumption across the training, deployment, and inference phases, embodying efficiency in every step of AI model lifecycle.

Simultaneously, we recognize the intrinsic link between software and hardware, advocating for a holistic co-design approach. This synergy ensures that our energy-efficient AI models are seamlessly integrated with hardware specifically engineered to support these models, enhancing their performance while minimizing power requirements. This synergy isn’t just about making individual components energy-efficient; it’s about creating a harmonious ecosystem where software and hardware mutually amplify their energy-saving capabilities. People use equipment or devices to interact with and understand the world, but we are often constrained by the computational resources or battery life, e.g., in the case of drones, robots, and autonomous vehicles. By embedding our optimized AI models into these devices, we empower them to perform complex tasks with precision and intelligence, while ensuring that every joule of energy is judiciously utilized. Green computing ensures that as our devices become more integrated into our lives, they do so with the lightest possible impact on our planet.

Green Computing – Research themes

  1. Green Algorithms: The focus here is on creating smarter, energy-efficient AI algorithms. We’re asking how to build powerful AI that uses less energy, aiming for models that are both high-performing and eco-friendly. Our aim is to set new standards for efficiency, where less energy delivers more intelligence.
  2. Green Hardware: In this theme, we explore the physical side of AI. We aim to design advanced, low-energy AI computation architectures that redefine the current energy use. The goal is to make hardware that’s not only powered by clean energy sources but also more energy-efficient for AI computations, pushing the boundaries of what’s possible in sustainable technology.
  3. Green Applications: Applying our green algorithms and hardware, we’re looking at how AI can be integrated into everyday applications to make them more sustainable. Our aim is to adapt AI for use in energy-sensitive settings and use it to improve the energy efficiency of systems across various sectors, making everyday technologies greener.

Want to know more?

The Faculty of Engineering’s Net-Zero Research

The School of Computer Science

Team leader

Chang Xu