Qi Sun Receives NSF CAREER Award to Design More Efficient AI Imagery
Associate Professor of Computer Science and Engineering Qi Sun has received a National Science Foundation CAREER Award to support research aimed at making generative artificial intelligence dramatically more efficient by aligning computing demands with the limits of human visual perception.
The NSF CAREER Award, one of the foundation’s highest honors for early-career faculty, will support Sun’s effort to rethink how AI systems generate images and video. While today’s generative AI models can produce increasingly realistic visual content, they also require enormous amounts of electricity, memory, and computing infrastructure. Sun’s project asks a deceptively simple question: how much of that computation is actually necessary for people to perceive high-quality images?
The research seeks to bridge the gap between the immense computational and energy expense of generative AI systems and the comparatively limited amount of visual information the human brain can process at any one time. By tailoring AI systems to human perception, Sun hopes to develop methods that reduce power consumption and operational costs while speeding up content generation.
“Current generative AI systems often devote resources to visual details that people may not even notice,” said Sun, who is also a member of the Center for Urban Science + Progress. “This project is about understanding the relationship between computing power cost and perceived image quality so we can build large-scale AI infrastructures, such as the emerging data centers, that are both more efficient and more human-centered.”
To accomplish this, the project will combine psychophysical studies of human vision with the creation of large-scale datasets that measure how hardware costs relate to perceived visual quality. Sun will also develop new probabilistic AI models and adaptive computing frameworks designed to maximize image quality under strict memory and energy constraints.
A central component of the work involves creating a shared computing module capable of dynamically allocating computational resources where they matter most perceptually. The research team will test the approach through a case study in urban planning conducted in partnership with small businesses, demonstrating how resource-aware generative AI could be deployed in real-world settings.
“As generative AI continues to grow, it is imperative to understand how it best functions for people while offsetting potential energy costs,” said Juan de Pablo, Anne and Joel Ehrenkranz Executive Vice President for Global Science and Technology at New York University and Executive Dean of NYU Tandon. “This CAREER Award will ensure that Qi Sun’s work continues to inform the development of smarter, faster, and cleaner AI models that work for users’ needs.”
Beyond advancing computer graphics and AI, the project addresses growing concerns about the environmental and economic costs of large-scale AI systems. As companies race to build ever larger models, the electricity demands of data centers and AI infrastructure have surged. Sun’s work aims to show that smarter, perception-driven design could deliver high-quality results without requiring ever-increasing computational power.
