Energy Efficient AI & Machine Learning

Artificial intelligence is rapidly reshaping computing, communications, science, and society, but the growing scale of AI training and inference is also driving unprecedented energy demand. Various IEE researchers are developing the next generation of energy-efficient AI systems, algorithms, and hardware to dramatically reduce the computational and environmental costs of machine learning across the cloud, data center, and edge. This research theme spans the full AI stack, from scalable machine learning algorithms and efficient model architectures to specialized hardware designed for low-power inference and optimization. Faculty researchers are advancing methods in natural language processing, reasoning, probabilistic computing, and large-scale AI systems that improve performance while reducing energy consumption, memory access, and computational overhead. Researchers are also developing benchmarking platforms and systems-level optimization tools that evaluate the energy efficiency of modern AI workflows, including model training, fine-tuning, and inference. These efforts provide practical frameworks for comparing AI models and hardware platforms while helping guide the development of more energy-efficient AI technologies.

At the hardware level, the initiative explores unconventional computing architectures inspired by physics and the human brain, including probabilistic computing, neuromorphic systems, and mixed-signal circuits that leverage stochastic behavior for efficient optimization and machine learning. Novel approaches to sparse and low-complexity AI inference are reducing memory bandwidth demands and minimizing energy-intensive operations in large language models and other advanced AI systems.Together, these efforts aim to enable scalable, high-performance AI technologies that are substantially more energy efficient, supporting the future of intelligent computing while simultaneously reducing its global energy footprint.

Lead Faculty